1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright 2010 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 /* 28 * tavor_cfg.c 29 * Tavor Configuration Profile Routines 30 * 31 * Implements the routines necessary for initializing and (later) tearing 32 * down the list of Tavor configuration information. 33 */ 34 35 #include <sys/types.h> 36 #include <sys/conf.h> 37 #include <sys/ddi.h> 38 #include <sys/sunddi.h> 39 #include <sys/modctl.h> 40 #include <sys/bitmap.h> 41 42 #include <sys/ib/adapters/tavor/tavor.h> 43 44 /* Set to enable alternative configurations: 0 = automatic config, 1 = manual */ 45 uint32_t tavor_alt_config_enable = 0; 46 47 /* Number of supported QPs and their maximum size */ 48 uint32_t tavor_log_num_qp = TAVOR_NUM_QP_SHIFT_128; 49 uint32_t tavor_log_max_qp_sz = TAVOR_QP_SZ_SHIFT; 50 51 /* Number of supported SGL per WQE */ 52 uint32_t tavor_wqe_max_sgl = TAVOR_NUM_WQE_SGL; 53 54 /* Number of supported CQs and their maximum size */ 55 uint32_t tavor_log_num_cq = TAVOR_NUM_CQ_SHIFT_128; 56 uint32_t tavor_log_max_cq_sz = TAVOR_CQ_SZ_SHIFT; 57 58 /* Select to enable SRQ or not; NOTE: 0 for disabled, 1 for enabled */ 59 uint32_t tavor_srq_enable = 1; 60 61 /* Number of supported SRQs and their maximum size */ 62 uint32_t tavor_log_num_srq = TAVOR_NUM_SRQ_SHIFT_128; 63 uint32_t tavor_log_max_srq_sz = TAVOR_SRQ_SZ_SHIFT; 64 uint32_t tavor_srq_max_sgl = TAVOR_SRQ_MAX_SGL; 65 66 /* Default size for all EQs */ 67 uint32_t tavor_log_default_eq_sz = TAVOR_DEFAULT_EQ_SZ_SHIFT; 68 69 /* Number of supported RDB (for incoming RDMA Read/Atomic) */ 70 uint32_t tavor_log_num_rdb = TAVOR_NUM_RDB_SHIFT_128; 71 72 /* 73 * Number of support multicast groups, number of QP per multicast group, and 74 * the number of entries (from the total number) in the multicast group "hash 75 * table" 76 */ 77 uint32_t tavor_log_num_mcg = TAVOR_NUM_MCG_SHIFT; 78 uint32_t tavor_num_qp_per_mcg = TAVOR_NUM_QP_PER_MCG; 79 uint32_t tavor_log_num_mcg_hash = TAVOR_NUM_MCG_HASH_SHIFT; 80 81 /* 82 * Number of supported MPTs (memory regions and windows) and their maximum 83 * size. Also the number of MTT per "MTT segment" (see tavor_mr.h for more 84 * details) 85 */ 86 uint32_t tavor_log_num_mpt = TAVOR_NUM_MPT_SHIFT_128; 87 uint32_t tavor_log_max_mrw_sz = TAVOR_MAX_MEM_MPT_SHIFT_128; 88 uint32_t tavor_log_num_mttseg = TAVOR_NUM_MTTSEG_SHIFT; 89 90 /* 91 * Number of supported Tavor mailboxes ("In" and "Out") and their maximum 92 * sizes, respectively 93 */ 94 uint32_t tavor_log_num_inmbox = TAVOR_NUM_MAILBOXES_SHIFT; 95 uint32_t tavor_log_num_outmbox = TAVOR_NUM_MAILBOXES_SHIFT; 96 uint32_t tavor_log_num_intr_inmbox = TAVOR_NUM_INTR_MAILBOXES_SHIFT; 97 uint32_t tavor_log_num_intr_outmbox = TAVOR_NUM_INTR_MAILBOXES_SHIFT; 98 uint32_t tavor_log_inmbox_size = TAVOR_MBOX_SIZE_SHIFT; 99 uint32_t tavor_log_outmbox_size = TAVOR_MBOX_SIZE_SHIFT; 100 101 /* Number of supported UAR pages */ 102 uint32_t tavor_log_num_uar = TAVOR_NUM_UAR_SHIFT; 103 104 /* Number of supported Protection Domains (PD) */ 105 uint32_t tavor_log_num_pd = TAVOR_NUM_PD_SHIFT; 106 107 /* Number of supported Address Handles (AH) */ 108 uint32_t tavor_log_num_ah = TAVOR_NUM_AH_SHIFT; 109 110 /* 111 * Number of total supported PKeys per PKey table (i.e. 112 * per port). Also the number of SGID per GID table. 113 */ 114 uint32_t tavor_log_max_pkeytbl = TAVOR_NUM_PKEYTBL_SHIFT; 115 uint32_t tavor_log_max_gidtbl = TAVOR_NUM_GIDTBL_SHIFT; 116 117 /* Maximum "responder resources" (in) and "initiator depth" (out) per QP */ 118 uint32_t tavor_hca_max_rdma_in_qp = TAVOR_HCA_MAX_RDMA_IN_QP; 119 uint32_t tavor_hca_max_rdma_out_qp = TAVOR_HCA_MAX_RDMA_OUT_QP; 120 121 /* Maximum supported MTU and portwidth */ 122 uint32_t tavor_max_mtu = TAVOR_MAX_MTU; 123 uint32_t tavor_max_port_width = TAVOR_MAX_PORT_WIDTH; 124 125 /* Number of supported Virtual Lanes (VL) */ 126 uint32_t tavor_max_vlcap = TAVOR_MAX_VLCAP; 127 128 /* Number of supported ports (1 or 2) */ 129 uint32_t tavor_num_ports = TAVOR_NUM_PORTS; 130 131 /* 132 * Whether or not to use the built-in (i.e. in firmware) agents for QP0 and 133 * QP1, respectively. 134 */ 135 uint32_t tavor_qp0_agents_in_fw = 1; 136 uint32_t tavor_qp1_agents_in_fw = 0; 137 138 /* 139 * Whether DMA mappings should be made with DDI_DMA_STREAMING or with 140 * DDI_DMA_CONSISTENT mode. Note: 0 for "streaming", 1 for "consistent" 141 */ 142 uint32_t tavor_streaming_consistent = 1; 143 144 /* 145 * For DMA mappings made with DDI_DMA_CONSISTENT, this flag determines 146 * whether to override the necessity for calls to ddi_dma_sync(). 147 */ 148 uint32_t tavor_consistent_syncoverride = 0; 149 150 /* 151 * Whether DMA mappings should bypass the PCI IOMMU or not. 152 * tavor_iommu_bypass is a global setting for all memory addresses. However, 153 * if set to BYPASS, memory attempted to be registered for streaming (ie: 154 * NON-COHERENT) will necessarily turn off BYPASS for that registration. To 155 * instead disable streaming in this situation the 156 * 'tavor_disable_streaming_on_bypass' can be set to 1. This setting will 157 * change the memory mapping to be implicitly consistent (ie: COHERENT), and 158 * will still perform the iommu BYPASS operation. 159 */ 160 uint32_t tavor_iommu_bypass = 1; 161 uint32_t tavor_disable_streaming_on_bypass = 0; 162 163 /* 164 * Whether QP work queues should be allocated from system memory or 165 * from Tavor DDR memory. Note: 0 for system memory, 1 for DDR memory 166 */ 167 uint32_t tavor_qp_wq_inddr = 0; 168 169 /* 170 * Whether SRQ work queues should be allocated from system memory or 171 * from Tavor DDR memory. Note: 0 for system memory, 1 for DDR memory 172 */ 173 uint32_t tavor_srq_wq_inddr = 0; 174 175 /* 176 * Whether Tavor should use MSI (Message Signaled Interrupts), if available. 177 * Note: 0 indicates 'legacy interrupt', 1 indicates MSI (if available) 178 */ 179 uint32_t tavor_use_msi_if_avail = 1; 180 181 /* 182 * This is a patchable variable that determines the time we will wait after 183 * initiating SW reset before we do our first read from Tavor config space. 184 * If this value is set too small (less than the default 100ms), it is 185 * possible for Tavor hardware to be unready to respond to the config cycle 186 * reads. This could cause master abort on the PCI bridge. Note: If 187 * "tavor_sw_reset_delay" is set to zero, then no software reset of the Tavor 188 * device will be attempted. 189 */ 190 uint32_t tavor_sw_reset_delay = TAVOR_SW_RESET_DELAY; 191 192 /* 193 * These are patchable variables for tavor command polling. The poll_delay is 194 * the number of usec to wait in-between calls to poll the 'go' bit. The 195 * poll_max is the total number of usec to loop in waiting for the 'go' bit to 196 * clear. 197 */ 198 uint32_t tavor_cmd_poll_delay = TAVOR_CMD_POLL_DELAY; 199 uint32_t tavor_cmd_poll_max = TAVOR_CMD_POLL_MAX; 200 201 /* 202 * This is a patchable variable that determines the frequency with which 203 * the AckReq bit will be set in outgoing RC packets. The AckReq bit will be 204 * set in at least every 2^tavor_qp_ackreq_freq packets (but at least once 205 * per message, i.e. in the last packet). Tuning this value can increase 206 * IB fabric utilization by cutting down on the number of unnecessary ACKs. 207 */ 208 uint32_t tavor_qp_ackreq_freq = TAVOR_QP_ACKREQ_FREQ; 209 210 /* 211 * This is a patchable variable that determines the default value for the 212 * maximum number of outstanding split transactions. The number of 213 * outstanding split transations (i.e. PCI reads) has an affect on device 214 * throughput. The value here should not be modified as it defines the 215 * default (least common denominator - one (1) PCI read) behavior that is 216 * guaranteed to work, regardless of how the Tavor firmware has been 217 * initialized. The format for this variable is the same as the corresponding 218 * field in the "PCI-X Command Register". 219 */ 220 #ifdef __sparc 221 /* 222 * Default SPARC platforms to be 1 outstanding PCI read. 223 */ 224 int tavor_max_out_splt_trans = 0; 225 #else 226 /* 227 * Default non-SPARC platforms to be the default as set in tavor firmware 228 * number of outstanding PCI reads. 229 */ 230 int tavor_max_out_splt_trans = -1; 231 #endif 232 233 /* 234 * This is a patchable variable that determines the default value for the 235 * maximum size of PCI read burst. This maximum size has an affect on 236 * device throughput. The value here should not be modified as it defines 237 * the default (least common denominator - 512B read) behavior that is 238 * guaranteed to work, regardless of how the Tavor device has been 239 * initialized. The format for this variable is the same as the corresponding 240 * field in the "PCI-X Command Register". 241 */ 242 #ifdef __sparc 243 /* 244 * Default SPARC platforms to be 512B read. 245 */ 246 int tavor_max_mem_rd_byte_cnt = 0; 247 static void tavor_check_iommu_bypass(tavor_state_t *state, 248 tavor_cfg_profile_t *cp); 249 #else 250 /* 251 * Default non-SPARC platforms to be the default as set in tavor firmware. 252 * 253 */ 254 int tavor_max_mem_rd_byte_cnt = -1; 255 #endif 256 257 static void tavor_cfg_wqe_sizes(tavor_cfg_profile_t *cp); 258 static void tavor_cfg_prop_lookup(tavor_state_t *state, 259 tavor_cfg_profile_t *cp); 260 261 /* 262 * tavor_cfg_profile_init_phase1() 263 * Context: Only called from attach() path context 264 */ 265 int 266 tavor_cfg_profile_init_phase1(tavor_state_t *state) 267 { 268 tavor_cfg_profile_t *cp; 269 270 TAVOR_TNF_ENTER(tavor_cfg_profile_init_phase1); 271 272 /* 273 * Allocate space for the configuration profile structure 274 */ 275 cp = (tavor_cfg_profile_t *)kmem_zalloc(sizeof (tavor_cfg_profile_t), 276 KM_SLEEP); 277 278 cp->cp_qp0_agents_in_fw = tavor_qp0_agents_in_fw; 279 cp->cp_qp1_agents_in_fw = tavor_qp1_agents_in_fw; 280 cp->cp_sw_reset_delay = tavor_sw_reset_delay; 281 cp->cp_cmd_poll_delay = tavor_cmd_poll_delay; 282 cp->cp_cmd_poll_max = tavor_cmd_poll_max; 283 cp->cp_ackreq_freq = tavor_qp_ackreq_freq; 284 cp->cp_max_out_splt_trans = tavor_max_out_splt_trans; 285 cp->cp_max_mem_rd_byte_cnt = tavor_max_mem_rd_byte_cnt; 286 cp->cp_srq_enable = tavor_srq_enable; 287 cp->cp_fmr_enable = 0; 288 cp->cp_fmr_max_remaps = 0; 289 290 /* 291 * Although most of the configuration is enabled in "phase2" of the 292 * cfg_profile_init, we have to setup the OUT mailboxes here, since 293 * they are used immediately after this "phase1" completes. Check for 294 * alt_config_enable, and set the values appropriately. Otherwise, the 295 * config profile is setup using the values based on the dimm size. 296 * While it is expected that the mailbox size and number will remain 297 * the same independent of dimm size, we separate it out here anyway 298 * for completeness. 299 * 300 * We have to setup SRQ settings here because MOD_STAT_CFG must be 301 * called before our call to QUERY_DEVLIM. If SRQ is enabled, then we 302 * must enable it in the firmware so that the phase2 settings will have 303 * the right device limits. 304 */ 305 if (tavor_alt_config_enable) { 306 cp->cp_log_num_outmbox = tavor_log_num_outmbox; 307 cp->cp_log_num_intr_outmbox = tavor_log_num_intr_outmbox; 308 cp->cp_log_outmbox_size = tavor_log_outmbox_size; 309 cp->cp_log_num_inmbox = tavor_log_num_inmbox; 310 cp->cp_log_num_intr_inmbox = tavor_log_num_intr_inmbox; 311 cp->cp_log_inmbox_size = tavor_log_inmbox_size; 312 cp->cp_log_num_srq = tavor_log_num_srq; 313 cp->cp_log_max_srq_sz = tavor_log_max_srq_sz; 314 315 } else if (state->ts_cfg_profile_setting >= TAVOR_DDR_SIZE_256) { 316 cp->cp_log_num_outmbox = TAVOR_NUM_MAILBOXES_SHIFT; 317 cp->cp_log_num_intr_outmbox = 318 TAVOR_NUM_INTR_MAILBOXES_SHIFT; 319 cp->cp_log_outmbox_size = TAVOR_MBOX_SIZE_SHIFT; 320 cp->cp_log_num_inmbox = TAVOR_NUM_MAILBOXES_SHIFT; 321 cp->cp_log_num_intr_inmbox = 322 TAVOR_NUM_INTR_MAILBOXES_SHIFT; 323 cp->cp_log_inmbox_size = TAVOR_MBOX_SIZE_SHIFT; 324 cp->cp_log_num_srq = TAVOR_NUM_SRQ_SHIFT_256; 325 cp->cp_log_max_srq_sz = TAVOR_SRQ_SZ_SHIFT; 326 327 } else if (state->ts_cfg_profile_setting == TAVOR_DDR_SIZE_128) { 328 cp->cp_log_num_outmbox = TAVOR_NUM_MAILBOXES_SHIFT; 329 cp->cp_log_num_intr_outmbox = 330 TAVOR_NUM_INTR_MAILBOXES_SHIFT; 331 cp->cp_log_outmbox_size = TAVOR_MBOX_SIZE_SHIFT; 332 cp->cp_log_num_inmbox = TAVOR_NUM_MAILBOXES_SHIFT; 333 cp->cp_log_num_intr_inmbox = 334 TAVOR_NUM_INTR_MAILBOXES_SHIFT; 335 cp->cp_log_inmbox_size = TAVOR_MBOX_SIZE_SHIFT; 336 cp->cp_log_num_srq = TAVOR_NUM_SRQ_SHIFT_128; 337 cp->cp_log_max_srq_sz = TAVOR_SRQ_SZ_SHIFT; 338 339 } else if (state->ts_cfg_profile_setting == TAVOR_DDR_SIZE_MIN) { 340 cp->cp_log_num_outmbox = TAVOR_NUM_MAILBOXES_SHIFT; 341 cp->cp_log_num_intr_outmbox = 342 TAVOR_NUM_INTR_MAILBOXES_SHIFT; 343 cp->cp_log_outmbox_size = TAVOR_MBOX_SIZE_SHIFT; 344 cp->cp_log_num_inmbox = TAVOR_NUM_MAILBOXES_SHIFT; 345 cp->cp_log_num_intr_inmbox = 346 TAVOR_NUM_INTR_MAILBOXES_SHIFT; 347 cp->cp_log_inmbox_size = TAVOR_MBOX_SIZE_SHIFT; 348 cp->cp_log_num_srq = TAVOR_NUM_SRQ_SHIFT_MIN; 349 cp->cp_log_max_srq_sz = TAVOR_SRQ_SZ_SHIFT_MIN; 350 351 } else { 352 TNF_PROBE_0(tavor_cfg_profile_invalid_dimmsz_fail, 353 TAVOR_TNF_ERROR, ""); 354 return (DDI_FAILURE); 355 } 356 357 /* 358 * Set default DMA mapping mode. Ensure consistency of flags 359 * with both architecture type and other configuration flags. 360 */ 361 if (tavor_streaming_consistent == 0) { 362 #ifdef __sparc 363 cp->cp_streaming_consistent = DDI_DMA_STREAMING; 364 365 /* Can't do both "streaming" and IOMMU bypass */ 366 if (tavor_iommu_bypass != 0) { 367 TNF_PROBE_0(tavor_cfg_profile_streamingbypass_fail, 368 TAVOR_TNF_ERROR, ""); 369 kmem_free(cp, sizeof (tavor_cfg_profile_t)); 370 return (DDI_FAILURE); 371 } 372 #else 373 cp->cp_streaming_consistent = DDI_DMA_CONSISTENT; 374 #endif 375 } else { 376 cp->cp_streaming_consistent = DDI_DMA_CONSISTENT; 377 } 378 379 /* Determine whether to override ddi_dma_sync() */ 380 cp->cp_consistent_syncoverride = tavor_consistent_syncoverride; 381 382 /* Attach the configuration profile to Tavor softstate */ 383 state->ts_cfg_profile = cp; 384 385 TAVOR_TNF_EXIT(tavor_cfg_profile_init_phase1); 386 return (DDI_SUCCESS); 387 } 388 389 /* 390 * tavor_cfg_profile_init_phase2() 391 * Context: Only called from attach() path context 392 */ 393 int 394 tavor_cfg_profile_init_phase2(tavor_state_t *state) 395 { 396 tavor_cfg_profile_t *cp; 397 398 TAVOR_TNF_ENTER(tavor_cfg_profile_init_phase2); 399 400 /* Read the configuration profile from Tavor softstate */ 401 cp = state->ts_cfg_profile; 402 403 /* 404 * Verify the config profile setting. The 'setting' should already be 405 * set, during a call to ddi_dev_regsize() to get the size of DDR 406 * memory, or during a fallback to a smaller supported size. If it is 407 * not set, we should not have reached this 'phase2'. So we assert 408 * here. 409 */ 410 ASSERT(state->ts_cfg_profile_setting != 0); 411 412 /* 413 * The automatic configuration override is the 414 * 'tavor_alt_config_enable' variable. If this is set, we no longer 415 * use the DIMM size to enable the correct profile. Instead, all of 416 * the tavor config options at the top of this file are used directly. 417 * 418 * This allows customization for a user who knows what they are doing 419 * to set tavor configuration values manually. 420 * 421 * If this variable is 0, we do automatic config for both 128MB and 422 * 256MB DIMM sizes. 423 */ 424 if (tavor_alt_config_enable) { 425 /* 426 * Initialize the configuration profile 427 */ 428 cp->cp_log_num_qp = tavor_log_num_qp; 429 cp->cp_log_max_qp_sz = tavor_log_max_qp_sz; 430 431 /* Determine WQE sizes from requested max SGLs */ 432 tavor_cfg_wqe_sizes(cp); 433 434 cp->cp_log_num_cq = tavor_log_num_cq; 435 cp->cp_log_max_cq_sz = tavor_log_max_cq_sz; 436 cp->cp_log_default_eq_sz = tavor_log_default_eq_sz; 437 cp->cp_log_num_rdb = tavor_log_num_rdb; 438 cp->cp_log_num_mcg = tavor_log_num_mcg; 439 cp->cp_num_qp_per_mcg = tavor_num_qp_per_mcg; 440 cp->cp_log_num_mcg_hash = tavor_log_num_mcg_hash; 441 cp->cp_log_num_mpt = tavor_log_num_mpt; 442 cp->cp_log_max_mrw_sz = tavor_log_max_mrw_sz; 443 cp->cp_log_num_mttseg = tavor_log_num_mttseg; 444 cp->cp_log_num_uar = tavor_log_num_uar; 445 cp->cp_log_num_pd = tavor_log_num_pd; 446 cp->cp_log_num_ah = tavor_log_num_ah; 447 cp->cp_log_max_pkeytbl = tavor_log_max_pkeytbl; 448 cp->cp_log_max_gidtbl = tavor_log_max_gidtbl; 449 cp->cp_hca_max_rdma_in_qp = tavor_hca_max_rdma_in_qp; 450 cp->cp_hca_max_rdma_out_qp = tavor_hca_max_rdma_out_qp; 451 cp->cp_max_mtu = tavor_max_mtu; 452 cp->cp_max_port_width = tavor_max_port_width; 453 cp->cp_max_vlcap = tavor_max_vlcap; 454 cp->cp_num_ports = tavor_num_ports; 455 cp->cp_qp0_agents_in_fw = tavor_qp0_agents_in_fw; 456 cp->cp_qp1_agents_in_fw = tavor_qp1_agents_in_fw; 457 cp->cp_sw_reset_delay = tavor_sw_reset_delay; 458 cp->cp_ackreq_freq = tavor_qp_ackreq_freq; 459 cp->cp_max_out_splt_trans = tavor_max_out_splt_trans; 460 cp->cp_max_mem_rd_byte_cnt = tavor_max_mem_rd_byte_cnt; 461 462 } else if (state->ts_cfg_profile_setting >= TAVOR_DDR_SIZE_256) { 463 /* 464 * Initialize the configuration profile 465 */ 466 cp->cp_log_num_qp = TAVOR_NUM_QP_SHIFT_256; 467 cp->cp_log_max_qp_sz = TAVOR_QP_SZ_SHIFT; 468 469 /* Determine WQE sizes from requested max SGLs */ 470 tavor_cfg_wqe_sizes(cp); 471 472 cp->cp_log_num_cq = TAVOR_NUM_CQ_SHIFT_256; 473 cp->cp_log_max_cq_sz = TAVOR_CQ_SZ_SHIFT; 474 cp->cp_log_default_eq_sz = TAVOR_DEFAULT_EQ_SZ_SHIFT; 475 cp->cp_log_num_rdb = TAVOR_NUM_RDB_SHIFT_256; 476 cp->cp_log_num_mcg = TAVOR_NUM_MCG_SHIFT; 477 cp->cp_num_qp_per_mcg = TAVOR_NUM_QP_PER_MCG; 478 cp->cp_log_num_mcg_hash = TAVOR_NUM_MCG_HASH_SHIFT; 479 cp->cp_log_num_mpt = TAVOR_NUM_MPT_SHIFT_256; 480 cp->cp_log_max_mrw_sz = TAVOR_MAX_MEM_MPT_SHIFT_256; 481 cp->cp_log_num_mttseg = TAVOR_NUM_MTTSEG_SHIFT; 482 cp->cp_log_num_uar = TAVOR_NUM_UAR_SHIFT; 483 cp->cp_log_num_pd = TAVOR_NUM_PD_SHIFT; 484 cp->cp_log_num_ah = TAVOR_NUM_AH_SHIFT; 485 cp->cp_log_max_pkeytbl = TAVOR_NUM_PKEYTBL_SHIFT; 486 cp->cp_log_max_gidtbl = TAVOR_NUM_GIDTBL_SHIFT; 487 cp->cp_hca_max_rdma_in_qp = TAVOR_HCA_MAX_RDMA_IN_QP; 488 cp->cp_hca_max_rdma_out_qp = TAVOR_HCA_MAX_RDMA_OUT_QP; 489 cp->cp_max_mtu = TAVOR_MAX_MTU; 490 cp->cp_max_port_width = TAVOR_MAX_PORT_WIDTH; 491 cp->cp_max_vlcap = TAVOR_MAX_VLCAP; 492 cp->cp_num_ports = TAVOR_NUM_PORTS; 493 cp->cp_qp0_agents_in_fw = tavor_qp0_agents_in_fw; 494 cp->cp_qp1_agents_in_fw = tavor_qp1_agents_in_fw; 495 cp->cp_sw_reset_delay = tavor_sw_reset_delay; 496 cp->cp_ackreq_freq = tavor_qp_ackreq_freq; 497 cp->cp_max_out_splt_trans = tavor_max_out_splt_trans; 498 cp->cp_max_mem_rd_byte_cnt = tavor_max_mem_rd_byte_cnt; 499 500 } else if (state->ts_cfg_profile_setting == TAVOR_DDR_SIZE_128) { 501 /* 502 * Initialize the configuration profile 503 */ 504 cp->cp_log_num_qp = TAVOR_NUM_QP_SHIFT_128; 505 cp->cp_log_max_qp_sz = TAVOR_QP_SZ_SHIFT; 506 507 /* Determine WQE sizes from requested max SGLs */ 508 tavor_cfg_wqe_sizes(cp); 509 510 cp->cp_log_num_cq = TAVOR_NUM_CQ_SHIFT_128; 511 cp->cp_log_max_cq_sz = TAVOR_CQ_SZ_SHIFT; 512 cp->cp_log_default_eq_sz = TAVOR_DEFAULT_EQ_SZ_SHIFT; 513 cp->cp_log_num_rdb = TAVOR_NUM_RDB_SHIFT_128; 514 cp->cp_log_num_mcg = TAVOR_NUM_MCG_SHIFT; 515 cp->cp_num_qp_per_mcg = TAVOR_NUM_QP_PER_MCG; 516 cp->cp_log_num_mcg_hash = TAVOR_NUM_MCG_HASH_SHIFT; 517 cp->cp_log_num_mpt = TAVOR_NUM_MPT_SHIFT_128; 518 cp->cp_log_max_mrw_sz = TAVOR_MAX_MEM_MPT_SHIFT_128; 519 cp->cp_log_num_mttseg = TAVOR_NUM_MTTSEG_SHIFT; 520 cp->cp_log_num_uar = TAVOR_NUM_UAR_SHIFT; 521 cp->cp_log_num_pd = TAVOR_NUM_PD_SHIFT; 522 cp->cp_log_num_ah = TAVOR_NUM_AH_SHIFT; 523 cp->cp_log_max_pkeytbl = TAVOR_NUM_PKEYTBL_SHIFT; 524 cp->cp_log_max_gidtbl = TAVOR_NUM_GIDTBL_SHIFT; 525 cp->cp_hca_max_rdma_in_qp = TAVOR_HCA_MAX_RDMA_IN_QP; 526 cp->cp_hca_max_rdma_out_qp = TAVOR_HCA_MAX_RDMA_OUT_QP; 527 cp->cp_max_mtu = TAVOR_MAX_MTU; 528 cp->cp_max_port_width = TAVOR_MAX_PORT_WIDTH; 529 cp->cp_max_vlcap = TAVOR_MAX_VLCAP; 530 cp->cp_num_ports = TAVOR_NUM_PORTS; 531 cp->cp_qp0_agents_in_fw = tavor_qp0_agents_in_fw; 532 cp->cp_qp1_agents_in_fw = tavor_qp1_agents_in_fw; 533 cp->cp_sw_reset_delay = tavor_sw_reset_delay; 534 cp->cp_ackreq_freq = tavor_qp_ackreq_freq; 535 cp->cp_max_out_splt_trans = tavor_max_out_splt_trans; 536 cp->cp_max_mem_rd_byte_cnt = tavor_max_mem_rd_byte_cnt; 537 538 } else if (state->ts_cfg_profile_setting == TAVOR_DDR_SIZE_MIN) { 539 /* 540 * Initialize the configuration profile for minimal footprint. 541 */ 542 543 cp->cp_log_num_qp = TAVOR_NUM_QP_SHIFT_MIN; 544 cp->cp_log_max_qp_sz = TAVOR_QP_SZ_SHIFT_MIN; 545 546 /* Determine WQE sizes from requested max SGLs */ 547 tavor_cfg_wqe_sizes(cp); 548 549 cp->cp_log_num_cq = TAVOR_NUM_CQ_SHIFT_MIN; 550 cp->cp_log_max_cq_sz = TAVOR_CQ_SZ_SHIFT_MIN; 551 cp->cp_log_default_eq_sz = TAVOR_DEFAULT_EQ_SZ_SHIFT; 552 cp->cp_log_num_rdb = TAVOR_NUM_RDB_SHIFT_MIN; 553 cp->cp_log_num_mcg = TAVOR_NUM_MCG_SHIFT_MIN; 554 cp->cp_num_qp_per_mcg = TAVOR_NUM_QP_PER_MCG_MIN; 555 cp->cp_log_num_mcg_hash = TAVOR_NUM_MCG_HASH_SHIFT_MIN; 556 cp->cp_log_num_mpt = TAVOR_NUM_MPT_SHIFT_MIN; 557 cp->cp_log_max_mrw_sz = TAVOR_MAX_MEM_MPT_SHIFT_MIN; 558 cp->cp_log_num_mttseg = TAVOR_NUM_MTTSEG_SHIFT_MIN; 559 cp->cp_log_num_uar = TAVOR_NUM_UAR_SHIFT_MIN; 560 cp->cp_log_num_pd = TAVOR_NUM_PD_SHIFT; 561 cp->cp_log_num_ah = TAVOR_NUM_AH_SHIFT_MIN; 562 cp->cp_log_max_pkeytbl = TAVOR_NUM_PKEYTBL_SHIFT; 563 cp->cp_log_max_gidtbl = TAVOR_NUM_GIDTBL_SHIFT; 564 cp->cp_hca_max_rdma_in_qp = TAVOR_HCA_MAX_RDMA_IN_QP; 565 cp->cp_hca_max_rdma_out_qp = TAVOR_HCA_MAX_RDMA_OUT_QP; 566 cp->cp_max_mtu = TAVOR_MAX_MTU; 567 cp->cp_max_port_width = TAVOR_MAX_PORT_WIDTH; 568 cp->cp_max_vlcap = TAVOR_MAX_VLCAP; 569 cp->cp_num_ports = TAVOR_NUM_PORTS; 570 cp->cp_qp0_agents_in_fw = tavor_qp0_agents_in_fw; 571 cp->cp_qp1_agents_in_fw = tavor_qp1_agents_in_fw; 572 cp->cp_sw_reset_delay = tavor_sw_reset_delay; 573 cp->cp_ackreq_freq = tavor_qp_ackreq_freq; 574 cp->cp_max_out_splt_trans = tavor_max_out_splt_trans; 575 cp->cp_max_mem_rd_byte_cnt = tavor_max_mem_rd_byte_cnt; 576 577 } else { 578 TNF_PROBE_0(tavor_cfg_profile_invalid_dimmsz_fail, 579 TAVOR_TNF_ERROR, ""); 580 return (DDI_FAILURE); 581 } 582 583 /* 584 * Set IOMMU bypass or not. Ensure consistency of flags with 585 * architecture type. 586 */ 587 #ifdef __sparc 588 if (tavor_iommu_bypass == 1) { 589 tavor_check_iommu_bypass(state, cp); 590 } else { 591 cp->cp_iommu_bypass = TAVOR_BINDMEM_NORMAL; 592 cp->cp_disable_streaming_on_bypass = 0; 593 } 594 #else 595 cp->cp_iommu_bypass = TAVOR_BINDMEM_NORMAL; 596 cp->cp_disable_streaming_on_bypass = 0; 597 #endif 598 /* Set whether QP WQEs will be in DDR or not */ 599 cp->cp_qp_wq_inddr = (tavor_qp_wq_inddr == 0) ? 600 TAVOR_QUEUE_LOCATION_NORMAL : TAVOR_QUEUE_LOCATION_INDDR; 601 602 /* Set whether SRQ WQEs will be in DDR or not */ 603 cp->cp_srq_wq_inddr = (tavor_srq_wq_inddr == 0) ? 604 TAVOR_QUEUE_LOCATION_NORMAL : TAVOR_QUEUE_LOCATION_INDDR; 605 606 cp->cp_use_msi_if_avail = tavor_use_msi_if_avail; 607 608 /* Determine additional configuration from optional properties */ 609 tavor_cfg_prop_lookup(state, cp); 610 611 TAVOR_TNF_EXIT(tavor_cfg_profile_init_phase2); 612 return (DDI_SUCCESS); 613 } 614 615 616 /* 617 * tavor_cfg_profile_fini() 618 * Context: Only called from attach() and/or detach() path contexts 619 */ 620 void 621 tavor_cfg_profile_fini(tavor_state_t *state) 622 { 623 TAVOR_TNF_ENTER(tavor_cfg_profile_fini); 624 625 /* 626 * Free up the space for configuration profile 627 */ 628 kmem_free(state->ts_cfg_profile, sizeof (tavor_cfg_profile_t)); 629 630 TAVOR_TNF_EXIT(tavor_cfg_profile_fini); 631 } 632 633 634 /* 635 * tavor_cfg_wqe_sizes() 636 * Context: Only called from attach() path context 637 */ 638 static void 639 tavor_cfg_wqe_sizes(tavor_cfg_profile_t *cp) 640 { 641 uint_t max_size, log2; 642 uint_t max_sgl, real_max_sgl; 643 644 /* 645 * Get the requested maximum number SGL per WQE from the Tavor 646 * patchable variable 647 */ 648 max_sgl = tavor_wqe_max_sgl; 649 650 /* 651 * Use requested maximum number of SGL to calculate the max descriptor 652 * size (while guaranteeing that the descriptor size is a power-of-2 653 * cachelines). We have to use the calculation for QP1 MLX transport 654 * because the possibility that we might need to inline a GRH, along 655 * with all the other headers and alignment restrictions, sets the 656 * maximum for the number of SGLs that we can advertise support for. 657 */ 658 max_size = (TAVOR_QP_WQE_MLX_QP1_HDRS + (max_sgl << 4)); 659 log2 = highbit(max_size); 660 if ((max_size & (max_size - 1)) == 0) { 661 log2 = log2 - 1; 662 } 663 max_size = (1 << log2); 664 665 /* 666 * Now clip the maximum descriptor size based on Tavor HW maximum 667 */ 668 max_size = min(max_size, TAVOR_QP_WQE_MAX_SIZE); 669 670 /* 671 * Then use the calculated max descriptor size to determine the "real" 672 * maximum SGL (the number beyond which we would roll over to the next 673 * power-of-2). 674 */ 675 real_max_sgl = (max_size - TAVOR_QP_WQE_MLX_QP1_HDRS) >> 4; 676 677 /* Then save away this configuration information */ 678 cp->cp_wqe_max_sgl = max_sgl; 679 cp->cp_wqe_real_max_sgl = real_max_sgl; 680 681 /* SRQ SGL gets set to it's own patchable variable value */ 682 cp->cp_srq_max_sgl = tavor_srq_max_sgl; 683 } 684 685 686 /* 687 * tavor_cfg_prop_lookup() 688 * Context: Only called from attach() path context 689 */ 690 static void 691 tavor_cfg_prop_lookup(tavor_state_t *state, tavor_cfg_profile_t *cp) 692 { 693 uint_t num_ports, nelementsp; 694 uchar_t *datap; 695 int status; 696 697 /* 698 * Read the property defining the number of Tavor ports to 699 * support. If the property is undefined or invalid, then return. 700 * We return here assuming also that OBP is not supposed to be setting 701 * up other properties in this case (eg: HCA plugin cards). But if 702 * this property is valid, then we print out a message for the other 703 * properties to show an OBP error. 704 */ 705 num_ports = ddi_prop_get_int(DDI_DEV_T_ANY, state->ts_dip, 706 DDI_PROP_DONTPASS, "#ports", 0); 707 if ((num_ports > TAVOR_NUM_PORTS) || (num_ports == 0)) { 708 return; 709 } 710 cp->cp_num_ports = num_ports; 711 712 /* 713 * The system image guid is not currently supported in the 1275 714 * binding. So we leave this commented out for now. 715 */ 716 #ifdef SUPPORTED_IN_1275_BINDING 717 /* 718 * Read the property defining the value to use later to override the 719 * default SystemImageGUID (in firmware). If the property is 720 * undefined, then return. 721 */ 722 status = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, state->ts_dip, 723 DDI_PROP_DONTPASS, "system-image-guid", &datap, &nelementsp); 724 if (status == DDI_PROP_SUCCESS) { 725 cp->cp_sysimgguid = ((uint64_t *)datap)[0]; 726 ddi_prop_free(datap); 727 } else { 728 cmn_err(CE_NOTE, 729 "Unable to read OBP system-image-guid property"); 730 } 731 #endif 732 733 /* 734 * Read the property defining the value to use later to override 735 * the default SystemImageGUID (in firmware). If the property is 736 * undefined, then return. 737 */ 738 status = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, state->ts_dip, 739 DDI_PROP_DONTPASS, "node-guid", &datap, &nelementsp); 740 if (status == DDI_PROP_SUCCESS) { 741 cp->cp_nodeguid = ((uint64_t *)datap)[0]; 742 ddi_prop_free(datap); 743 } else { 744 cmn_err(CE_NOTE, "Unable to read OBP node-guid property"); 745 } 746 747 /* 748 * Using the value for the number of ports (above) read the properties 749 * used to later to override the default PortGUIDs for each Tavor port. 750 * If either of these properties are undefined, then return. 751 */ 752 if (num_ports == TAVOR_NUM_PORTS) { 753 status = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, 754 state->ts_dip, DDI_PROP_DONTPASS, "port-2-guid", &datap, 755 &nelementsp); 756 if (status == DDI_PROP_SUCCESS) { 757 cp->cp_portguid[1] = ((uint64_t *)datap)[0]; 758 ddi_prop_free(datap); 759 } else { 760 cmn_err(CE_NOTE, 761 "Unable to read OBP port-2-guid property"); 762 } 763 } 764 status = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, state->ts_dip, 765 DDI_PROP_DONTPASS, "port-1-guid", &datap, &nelementsp); 766 if (status == DDI_PROP_SUCCESS) { 767 cp->cp_portguid[0] = ((uint64_t *)datap)[0]; 768 ddi_prop_free(datap); 769 } else { 770 cmn_err(CE_NOTE, "Unable to read OBP port-1-guid property"); 771 } 772 } 773 774 #ifdef __sparc 775 /* 776 * tavor_check_iommu_bypass() 777 * Context: Only called from attach() path context 778 */ 779 static void 780 tavor_check_iommu_bypass(tavor_state_t *state, tavor_cfg_profile_t *cp) 781 { 782 ddi_dma_handle_t dmahdl; 783 ddi_dma_attr_t dma_attr; 784 int status; 785 786 tavor_dma_attr_init(&dma_attr); 787 788 /* Try mapping for IOMMU bypass (Force Physical) */ 789 dma_attr.dma_attr_flags = DDI_DMA_FORCE_PHYSICAL; 790 791 /* 792 * Call ddi_dma_alloc_handle(). If this returns DDI_DMA_BADATTR then 793 * it is not possible to use IOMMU bypass with our PCI bridge parent. 794 * For example, certain versions of Tomatillo do not support IOMMU 795 * bypass. Since the function we are in can only be called if iommu 796 * bypass was requested in the config profile, we configure for bypass 797 * if the ddi_dma_alloc_handle() was successful. Otherwise, we 798 * configure for non-bypass (ie: normal) mapping. 799 */ 800 status = ddi_dma_alloc_handle(state->ts_dip, &dma_attr, 801 DDI_DMA_SLEEP, NULL, &dmahdl); 802 if (status == DDI_DMA_BADATTR) { 803 cp->cp_iommu_bypass = TAVOR_BINDMEM_NORMAL; 804 cp->cp_disable_streaming_on_bypass = 0; 805 } else { 806 cp->cp_iommu_bypass = TAVOR_BINDMEM_BYPASS; 807 cp->cp_disable_streaming_on_bypass = 808 tavor_disable_streaming_on_bypass; 809 810 if (status == DDI_SUCCESS) { 811 ddi_dma_free_handle(&dmahdl); 812 } 813 } 814 } 815 #endif 816