1 /* 2 * CDDL HEADER START 3 * 4 * Copyright(c) 2007-2008 Intel Corporation. All rights reserved. 5 * The contents of this file are subject to the terms of the 6 * Common Development and Distribution License (the "License"). 7 * You may not use this file except in compliance with the License. 8 * 9 * You can obtain a copy of the license at: 10 * http://www.opensolaris.org/os/licensing. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When using or redistributing this file, you may do so under the 15 * License only. No other modification of this header is permitted. 16 * 17 * If applicable, add the following below this CDDL HEADER, with the 18 * fields enclosed by brackets "[]" replaced with your own identifying 19 * information: Portions Copyright [yyyy] [name of copyright owner] 20 * 21 * CDDL HEADER END 22 */ 23 24 /* 25 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 26 * Use is subject to license terms of the CDDL. 27 */ 28 29 #pragma ident "%Z%%M% %I% %E% SMI" 30 31 #include "ixgbe_sw.h" 32 33 /* function prototypes */ 34 static mblk_t *ixgbe_rx_bind(ixgbe_rx_ring_t *, uint32_t, uint32_t); 35 static mblk_t *ixgbe_rx_copy(ixgbe_rx_ring_t *, uint32_t, uint32_t); 36 static void ixgbe_rx_assoc_hcksum(mblk_t *, uint32_t); 37 38 #ifndef IXGBE_DEBUG 39 #pragma inline(ixgbe_rx_assoc_hcksum) 40 #endif 41 42 /* 43 * ixgbe_rx_recycle - The call-back function to reclaim rx buffer. 44 * 45 * This function is called when an mp is freed by the user thru 46 * freeb call (Only for mp constructed through desballoc call). 47 * It returns back the freed buffer to the free list. 48 */ 49 void 50 ixgbe_rx_recycle(caddr_t arg) 51 { 52 ixgbe_rx_ring_t *rx_ring; 53 rx_control_block_t *recycle_rcb; 54 uint32_t free_index; 55 56 recycle_rcb = (rx_control_block_t *)(uintptr_t)arg; 57 rx_ring = recycle_rcb->rx_ring; 58 59 if (recycle_rcb->state == RCB_FREE) 60 return; 61 62 recycle_rcb->state = RCB_FREE; 63 64 ASSERT(recycle_rcb->mp == NULL); 65 66 /* 67 * Using the recycled data buffer to generate a new mblk 68 */ 69 recycle_rcb->mp = desballoc((unsigned char *) 70 (recycle_rcb->rx_buf.address - IPHDR_ALIGN_ROOM), 71 (recycle_rcb->rx_buf.size + IPHDR_ALIGN_ROOM), 72 0, &recycle_rcb->free_rtn); 73 if (recycle_rcb->mp != NULL) { 74 recycle_rcb->mp->b_rptr += IPHDR_ALIGN_ROOM; 75 recycle_rcb->mp->b_wptr += IPHDR_ALIGN_ROOM; 76 } 77 78 /* 79 * Put the recycled rx control block into free list 80 */ 81 mutex_enter(&rx_ring->recycle_lock); 82 83 free_index = rx_ring->rcb_tail; 84 ASSERT(rx_ring->free_list[free_index] == NULL); 85 86 rx_ring->free_list[free_index] = recycle_rcb; 87 rx_ring->rcb_tail = NEXT_INDEX(free_index, 1, rx_ring->free_list_size); 88 89 mutex_exit(&rx_ring->recycle_lock); 90 91 /* 92 * The atomic operation on the number of the available rx control 93 * blocks in the free list is used to make the recycling mutual 94 * exclusive with the receiving. 95 */ 96 atomic_inc_32(&rx_ring->rcb_free); 97 ASSERT(rx_ring->rcb_free <= rx_ring->free_list_size); 98 } 99 100 /* 101 * ixgbe_rx_copy - Use copy to process the received packet. 102 * 103 * This function will use bcopy to process the packet 104 * and send the copied packet upstream. 105 */ 106 static mblk_t * 107 ixgbe_rx_copy(ixgbe_rx_ring_t *rx_ring, uint32_t index, uint32_t pkt_len) 108 { 109 rx_control_block_t *current_rcb; 110 mblk_t *mp; 111 112 current_rcb = rx_ring->work_list[index]; 113 114 DMA_SYNC(¤t_rcb->rx_buf, DDI_DMA_SYNC_FORKERNEL); 115 116 if (ixgbe_check_dma_handle(current_rcb->rx_buf.dma_handle) != 117 DDI_FM_OK) { 118 ddi_fm_service_impact(rx_ring->ixgbe->dip, 119 DDI_SERVICE_DEGRADED); 120 } 121 122 /* 123 * Allocate buffer to receive this packet 124 */ 125 mp = allocb(pkt_len + IPHDR_ALIGN_ROOM, 0); 126 if (mp == NULL) { 127 ixgbe_log(rx_ring->ixgbe, 128 "ixgbe_rx_copy: allocate buffer failed"); 129 return (NULL); 130 } 131 132 /* 133 * Copy the data received into the new cluster 134 */ 135 mp->b_rptr += IPHDR_ALIGN_ROOM; 136 bcopy(current_rcb->rx_buf.address, mp->b_rptr, pkt_len); 137 mp->b_wptr = mp->b_rptr + pkt_len; 138 139 return (mp); 140 } 141 142 /* 143 * ixgbe_rx_bind - Use existing DMA buffer to build mblk for receiving. 144 * 145 * This function will use pre-bound DMA buffer to receive the packet 146 * and build mblk that will be sent upstream. 147 */ 148 static mblk_t * 149 ixgbe_rx_bind(ixgbe_rx_ring_t *rx_ring, uint32_t index, uint32_t pkt_len) 150 { 151 rx_control_block_t *current_rcb; 152 rx_control_block_t *free_rcb; 153 uint32_t free_index; 154 mblk_t *mp; 155 156 /* 157 * If the free list is empty, we cannot proceed to send 158 * the current DMA buffer upstream. We'll have to return 159 * and use bcopy to process the packet. 160 */ 161 if (ixgbe_atomic_reserve(&rx_ring->rcb_free, 1) < 0) 162 return (NULL); 163 164 current_rcb = rx_ring->work_list[index]; 165 /* 166 * If the mp of the rx control block is NULL, try to do 167 * desballoc again. 168 */ 169 if (current_rcb->mp == NULL) { 170 current_rcb->mp = desballoc((unsigned char *) 171 (current_rcb->rx_buf.address - IPHDR_ALIGN_ROOM), 172 (current_rcb->rx_buf.size + IPHDR_ALIGN_ROOM), 173 0, ¤t_rcb->free_rtn); 174 /* 175 * If it is failed to built a mblk using the current 176 * DMA buffer, we have to return and use bcopy to 177 * process the packet. 178 */ 179 if (current_rcb->mp == NULL) { 180 atomic_inc_32(&rx_ring->rcb_free); 181 return (NULL); 182 } 183 } 184 /* 185 * Sync up the data received 186 */ 187 DMA_SYNC(¤t_rcb->rx_buf, DDI_DMA_SYNC_FORKERNEL); 188 189 if (ixgbe_check_dma_handle(current_rcb->rx_buf.dma_handle) != 190 DDI_FM_OK) { 191 ddi_fm_service_impact(rx_ring->ixgbe->dip, 192 DDI_SERVICE_DEGRADED); 193 } 194 195 mp = current_rcb->mp; 196 current_rcb->mp = NULL; 197 current_rcb->state = RCB_SENDUP; 198 199 mp->b_wptr = mp->b_rptr + pkt_len; 200 mp->b_next = mp->b_cont = NULL; 201 202 /* 203 * Strip off one free rx control block from the free list 204 */ 205 free_index = rx_ring->rcb_head; 206 free_rcb = rx_ring->free_list[free_index]; 207 ASSERT(free_rcb != NULL); 208 rx_ring->free_list[free_index] = NULL; 209 rx_ring->rcb_head = NEXT_INDEX(free_index, 1, rx_ring->free_list_size); 210 211 /* 212 * Put the rx control block to the work list 213 */ 214 rx_ring->work_list[index] = free_rcb; 215 216 return (mp); 217 } 218 219 /* 220 * ixgbe_rx_assoc_hcksum - Check the rx hardware checksum status and associate 221 * the hcksum flags. 222 */ 223 static void 224 ixgbe_rx_assoc_hcksum(mblk_t *mp, uint32_t status_error) 225 { 226 uint32_t hcksum_flags = 0; 227 228 /* 229 * Check TCP/UDP checksum 230 */ 231 if ((status_error & IXGBE_RXD_STAT_L4CS) && 232 !(status_error & IXGBE_RXDADV_ERR_TCPE)) 233 hcksum_flags |= HCK_FULLCKSUM | HCK_FULLCKSUM_OK; 234 235 /* 236 * Check IP Checksum 237 */ 238 if ((status_error & IXGBE_RXD_STAT_IPCS) && 239 !(status_error & IXGBE_RXDADV_ERR_IPE)) 240 hcksum_flags |= HCK_IPV4_HDRCKSUM; 241 242 if (hcksum_flags != 0) { 243 (void) hcksum_assoc(mp, 244 NULL, NULL, 0, 0, 0, 0, hcksum_flags, 0); 245 } 246 } 247 248 /* 249 * ixgbe_rx - Receive the data of one ring. 250 * 251 * This function goes throught h/w descriptor in one specified rx ring, 252 * receives the data if the descriptor status shows the data is ready. 253 * It returns a chain of mblks containing the received data, to be 254 * passed up to mac_rx(). 255 */ 256 mblk_t * 257 ixgbe_rx(ixgbe_rx_ring_t *rx_ring) 258 { 259 union ixgbe_adv_rx_desc *current_rbd; 260 rx_control_block_t *current_rcb; 261 mblk_t *mp; 262 mblk_t *mblk_head; 263 mblk_t **mblk_tail; 264 uint32_t rx_next; 265 uint32_t rx_tail; 266 uint32_t pkt_len; 267 uint32_t status_error; 268 uint32_t pkt_num; 269 ixgbe_t *ixgbe = rx_ring->ixgbe; 270 struct ixgbe_hw *hw = &ixgbe->hw; 271 272 mblk_head = NULL; 273 mblk_tail = &mblk_head; 274 275 /* 276 * Sync the receive descriptors before accepting the packets 277 */ 278 DMA_SYNC(&rx_ring->rbd_area, DDI_DMA_SYNC_FORKERNEL); 279 280 if (ixgbe_check_dma_handle(rx_ring->rbd_area.dma_handle) != DDI_FM_OK) { 281 ddi_fm_service_impact(rx_ring->ixgbe->dip, 282 DDI_SERVICE_DEGRADED); 283 } 284 285 /* 286 * Get the start point of rx bd ring which should be examined 287 * during this cycle. 288 */ 289 rx_next = rx_ring->rbd_next; 290 291 current_rbd = &rx_ring->rbd_ring[rx_next]; 292 pkt_num = 0; 293 status_error = current_rbd->wb.upper.status_error; 294 while (status_error & IXGBE_RXD_STAT_DD) { 295 /* 296 * If adapter has found errors, but the error 297 * is hardware checksum error, this does not discard the 298 * packet: let upper layer compute the checksum; 299 * Otherwise discard the packet. 300 */ 301 if ((status_error & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) || 302 !(status_error & IXGBE_RXD_STAT_EOP)) { 303 IXGBE_DEBUG_STAT(rx_ring->stat_frame_error); 304 goto rx_discard; 305 } 306 307 IXGBE_DEBUG_STAT_COND(rx_ring->stat_cksum_error, 308 (status_error & IXGBE_RXDADV_ERR_TCPE) || 309 (status_error & IXGBE_RXDADV_ERR_IPE)); 310 311 pkt_len = current_rbd->wb.upper.length; 312 mp = NULL; 313 /* 314 * For packets with length more than the copy threshold, 315 * we'll first try to use the existing DMA buffer to build 316 * an mblk and send the mblk upstream. 317 * 318 * If the first method fails, or the packet length is less 319 * than the copy threshold, we'll allocate a new mblk and 320 * copy the packet data to the new mblk. 321 */ 322 if (pkt_len > rx_ring->copy_thresh) 323 mp = ixgbe_rx_bind(rx_ring, rx_next, pkt_len); 324 325 if (mp == NULL) 326 mp = ixgbe_rx_copy(rx_ring, rx_next, pkt_len); 327 328 if (mp != NULL) { 329 /* 330 * Check h/w checksum offload status 331 */ 332 if (ixgbe->rx_hcksum_enable) 333 ixgbe_rx_assoc_hcksum(mp, status_error); 334 335 *mblk_tail = mp; 336 mblk_tail = &mp->b_next; 337 } 338 339 rx_discard: 340 /* 341 * Reset rx descriptor read bits 342 */ 343 current_rcb = rx_ring->work_list[rx_next]; 344 current_rbd->read.pkt_addr = current_rcb->rx_buf.dma_address; 345 current_rbd->read.hdr_addr = 0; 346 347 rx_next = NEXT_INDEX(rx_next, 1, rx_ring->ring_size); 348 349 /* 350 * The receive function is in interrupt context, so here 351 * limit_per_intr is used to avoid doing receiving too long 352 * per interrupt. 353 */ 354 if (++pkt_num > rx_ring->limit_per_intr) { 355 IXGBE_DEBUG_STAT(rx_ring->stat_exceed_pkt); 356 break; 357 } 358 359 current_rbd = &rx_ring->rbd_ring[rx_next]; 360 status_error = current_rbd->wb.upper.status_error; 361 } 362 363 DMA_SYNC(&rx_ring->rbd_area, DDI_DMA_SYNC_FORDEV); 364 365 rx_ring->rbd_next = rx_next; 366 367 /* 368 * Update the h/w tail accordingly 369 */ 370 rx_tail = PREV_INDEX(rx_next, 1, rx_ring->ring_size); 371 372 IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_RDT(rx_ring->index), rx_tail); 373 374 if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) { 375 ddi_fm_service_impact(rx_ring->ixgbe->dip, 376 DDI_SERVICE_DEGRADED); 377 } 378 379 return (mblk_head); 380 } 381