aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@g5.osdl.org>2006-09-29 18:18:22 -0400
committerLinus Torvalds <torvalds@g5.osdl.org>2006-09-29 18:18:22 -0400
commit9a69d1aeccf169d9a1e442c07d3a6e87f06a7b49 (patch)
tree5597011c3595867bf0e073b8f4bdffefe9238a10
parentc0341b0f47722fbe5ab45f436fc6ddc1c58c0a6f (diff)
parent3d27b00457167103fb9f7e23fc2454c801a6b8f0 (diff)
Merge branch 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband
* 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband: (33 commits) IB/ipath: Fix lockdep error upon "ifconfig ibN down" IB/ipath: Fix races with ib_resize_cq() IB/ipath: Support new PCIE device, QLE7142 IB/ipath: Set CPU affinity early IB/ipath: Fix EEPROM read when driver is compiled with -Os IB/ipath: Fix and recover TXE piobuf and PBC parity errors IB/ipath: Change HT CRC message to indicate how to resolve problem IB/ipath: Clean up module exit code IB/ipath: Call mtrr_del with correct arguments IB/ipath: Flush RWQEs if access error or invalid error seen IB/ipath: Improved support for PowerPC IB/ipath: Drop unnecessary "(void *)" casts IB/ipath: Support multiple simultaneous devices of different types IB/ipath: Fix mismatch in shifts and masks for printing debug info IB/ipath: Fix compiler warnings and errors on non-x86_64 systems IB/ipath: Print more informative parity error messages IB/ipath: Ensure that PD of MR matches PD of QP checking the Rkey IB/ipath: RC and UC should validate SLID and DLID IB/ipath: Only allow complete writes to flash IB/ipath: Count SRQs properly ...
Diffstat
-rw-r--r--drivers/infiniband/hw/amso1100/c2_ae.c2
-rw-r--r--drivers/infiniband/hw/amso1100/c2_alloc.c2
-rw-r--r--drivers/infiniband/hw/amso1100/c2_cm.c15
-rw-r--r--drivers/infiniband/hw/amso1100/c2_provider.c8
-rw-r--r--drivers/infiniband/hw/amso1100/c2_rnic.c4
-rw-r--r--drivers/infiniband/hw/ipath/ipath_common.h54
-rw-r--r--drivers/infiniband/hw/ipath/ipath_cq.c48
-rw-r--r--drivers/infiniband/hw/ipath/ipath_driver.c359
-rw-r--r--drivers/infiniband/hw/ipath/ipath_eeprom.c17
-rw-r--r--drivers/infiniband/hw/ipath/ipath_file_ops.c974
-rw-r--r--drivers/infiniband/hw/ipath/ipath_fs.c9
-rw-r--r--drivers/infiniband/hw/ipath/ipath_iba6110.c132
-rw-r--r--drivers/infiniband/hw/ipath/ipath_iba6120.c263
-rw-r--r--drivers/infiniband/hw/ipath/ipath_init_chip.c56
-rw-r--r--drivers/infiniband/hw/ipath/ipath_intr.c280
-rw-r--r--drivers/infiniband/hw/ipath/ipath_kernel.h116
-rw-r--r--drivers/infiniband/hw/ipath/ipath_keys.c12
-rw-r--r--drivers/infiniband/hw/ipath/ipath_mad.c16
-rw-r--r--drivers/infiniband/hw/ipath/ipath_mr.c3
-rw-r--r--drivers/infiniband/hw/ipath/ipath_qp.c16
-rw-r--r--drivers/infiniband/hw/ipath/ipath_rc.c77
-rw-r--r--drivers/infiniband/hw/ipath/ipath_registers.h40
-rw-r--r--drivers/infiniband/hw/ipath/ipath_ruc.c14
-rw-r--r--drivers/infiniband/hw/ipath/ipath_srq.c23
-rw-r--r--drivers/infiniband/hw/ipath/ipath_sysfs.c21
-rw-r--r--drivers/infiniband/hw/ipath/ipath_uc.c6
-rw-r--r--drivers/infiniband/hw/ipath/ipath_ud.c6
-rw-r--r--drivers/infiniband/hw/ipath/ipath_user_pages.c56
-rw-r--r--drivers/infiniband/hw/ipath/ipath_verbs.c43
-rw-r--r--drivers/infiniband/hw/ipath/ipath_verbs.h18
-rw-r--r--drivers/infiniband/hw/ipath/ipath_wc_ppc64.c20
-rw-r--r--drivers/infiniband/hw/ipath/ipath_wc_x86_64.c13
-rw-r--r--drivers/infiniband/ulp/iser/Kconfig13
-rw-r--r--drivers/infiniband/ulp/iser/iscsi_iser.c2
-rw-r--r--drivers/infiniband/ulp/iser/iscsi_iser.h9
-rw-r--r--drivers/infiniband/ulp/iser/iser_initiator.c60
-rw-r--r--drivers/infiniband/ulp/iser/iser_memory.c42
-rw-r--r--drivers/infiniband/ulp/iser/iser_verbs.c8
38 files changed, 1973 insertions, 884 deletions
diff --git a/drivers/infiniband/hw/amso1100/c2_ae.c b/drivers/infiniband/hw/amso1100/c2_ae.c
index 08f46c83a3a4..3aae4978e1cb 100644
--- a/drivers/infiniband/hw/amso1100/c2_ae.c
+++ b/drivers/infiniband/hw/amso1100/c2_ae.c
@@ -197,7 +197,7 @@ void c2_ae_event(struct c2_dev *c2dev, u32 mq_index)
197 "resource=%x, qp_state=%s\n", 197 "resource=%x, qp_state=%s\n",
198 __FUNCTION__, 198 __FUNCTION__,
199 to_event_str(event_id), 199 to_event_str(event_id),
200 be64_to_cpu(wr->ae.ae_generic.user_context), 200 (unsigned long long) be64_to_cpu(wr->ae.ae_generic.user_context),
201 be32_to_cpu(wr->ae.ae_generic.resource_type), 201 be32_to_cpu(wr->ae.ae_generic.resource_type),
202 be32_to_cpu(wr->ae.ae_generic.resource), 202 be32_to_cpu(wr->ae.ae_generic.resource),
203 to_qp_state_str(be32_to_cpu(wr->ae.ae_generic.qp_state))); 203 to_qp_state_str(be32_to_cpu(wr->ae.ae_generic.qp_state)));
diff --git a/drivers/infiniband/hw/amso1100/c2_alloc.c b/drivers/infiniband/hw/amso1100/c2_alloc.c
index 1d2529992c0c..028a60bbfca9 100644
--- a/drivers/infiniband/hw/amso1100/c2_alloc.c
+++ b/drivers/infiniband/hw/amso1100/c2_alloc.c
@@ -115,7 +115,7 @@ u16 *c2_alloc_mqsp(struct c2_dev *c2dev, struct sp_chunk *head,
115 ((unsigned long) &(head->shared_ptr[mqsp]) - 115 ((unsigned long) &(head->shared_ptr[mqsp]) -
116 (unsigned long) head); 116 (unsigned long) head);
117 pr_debug("%s addr %p dma_addr %llx\n", __FUNCTION__, 117 pr_debug("%s addr %p dma_addr %llx\n", __FUNCTION__,
118 &(head->shared_ptr[mqsp]), (u64)*dma_addr); 118 &(head->shared_ptr[mqsp]), (unsigned long long) *dma_addr);
119 return &(head->shared_ptr[mqsp]); 119 return &(head->shared_ptr[mqsp]);
120 } 120 }
121 return NULL; 121 return NULL;
diff --git a/drivers/infiniband/hw/amso1100/c2_cm.c b/drivers/infiniband/hw/amso1100/c2_cm.c
index 485254efdd1e..75b93e9b8810 100644
--- a/drivers/infiniband/hw/amso1100/c2_cm.c
+++ b/drivers/infiniband/hw/amso1100/c2_cm.c
@@ -302,7 +302,7 @@ int c2_llp_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param)
302 vq_req = vq_req_alloc(c2dev); 302 vq_req = vq_req_alloc(c2dev);
303 if (!vq_req) { 303 if (!vq_req) {
304 err = -ENOMEM; 304 err = -ENOMEM;
305 goto bail1; 305 goto bail0;
306 } 306 }
307 vq_req->qp = qp; 307 vq_req->qp = qp;
308 vq_req->cm_id = cm_id; 308 vq_req->cm_id = cm_id;
@@ -311,7 +311,7 @@ int c2_llp_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param)
311 wr = kmalloc(c2dev->req_vq.msg_size, GFP_KERNEL); 311 wr = kmalloc(c2dev->req_vq.msg_size, GFP_KERNEL);
312 if (!wr) { 312 if (!wr) {
313 err = -ENOMEM; 313 err = -ENOMEM;
314 goto bail2; 314 goto bail1;
315 } 315 }
316 316
317 /* Build the WR */ 317 /* Build the WR */
@@ -331,7 +331,7 @@ int c2_llp_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param)
331 /* Validate private_data length */ 331 /* Validate private_data length */
332 if (iw_param->private_data_len > C2_MAX_PRIVATE_DATA_SIZE) { 332 if (iw_param->private_data_len > C2_MAX_PRIVATE_DATA_SIZE) {
333 err = -EINVAL; 333 err = -EINVAL;
334 goto bail2; 334 goto bail1;
335 } 335 }
336 336
337 if (iw_param->private_data) { 337 if (iw_param->private_data) {
@@ -348,19 +348,19 @@ int c2_llp_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param)
348 err = vq_send_wr(c2dev, (union c2wr *) wr); 348 err = vq_send_wr(c2dev, (union c2wr *) wr);
349 if (err) { 349 if (err) {
350 vq_req_put(c2dev, vq_req); 350 vq_req_put(c2dev, vq_req);
351 goto bail2; 351 goto bail1;
352 } 352 }
353 353
354 /* Wait for reply from adapter */ 354 /* Wait for reply from adapter */
355 err = vq_wait_for_reply(c2dev, vq_req); 355 err = vq_wait_for_reply(c2dev, vq_req);
356 if (err) 356 if (err)
357 goto bail2; 357 goto bail1;
358 358
359 /* Check that reply is present */ 359 /* Check that reply is present */
360 reply = (struct c2wr_cr_accept_rep *) (unsigned long) vq_req->reply_msg; 360 reply = (struct c2wr_cr_accept_rep *) (unsigned long) vq_req->reply_msg;
361 if (!reply) { 361 if (!reply) {
362 err = -ENOMEM; 362 err = -ENOMEM;
363 goto bail2; 363 goto bail1;
364 } 364 }
365 365
366 err = c2_errno(reply); 366 err = c2_errno(reply);
@@ -368,9 +368,8 @@ int c2_llp_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param)
368 368
369 if (!err) 369 if (!err)
370 c2_set_qp_state(qp, C2_QP_STATE_RTS); 370 c2_set_qp_state(qp, C2_QP_STATE_RTS);
371 bail2:
372 kfree(wr);
373 bail1: 371 bail1:
372 kfree(wr);
374 vq_req_free(c2dev, vq_req); 373 vq_req_free(c2dev, vq_req);
375 bail0: 374 bail0:
376 if (err) { 375 if (err) {
diff --git a/drivers/infiniband/hw/amso1100/c2_provider.c b/drivers/infiniband/hw/amso1100/c2_provider.c
index dd6af551108b..da98d9f71429 100644
--- a/drivers/infiniband/hw/amso1100/c2_provider.c
+++ b/drivers/infiniband/hw/amso1100/c2_provider.c
@@ -390,14 +390,18 @@ static struct ib_mr *c2_reg_phys_mr(struct ib_pd *ib_pd,
390 } 390 }
391 391
392 mr = kmalloc(sizeof(*mr), GFP_KERNEL); 392 mr = kmalloc(sizeof(*mr), GFP_KERNEL);
393 if (!mr) 393 if (!mr) {
394 vfree(page_list);
394 return ERR_PTR(-ENOMEM); 395 return ERR_PTR(-ENOMEM);
396 }
395 397
396 mr->pd = to_c2pd(ib_pd); 398 mr->pd = to_c2pd(ib_pd);
397 pr_debug("%s - page shift %d, pbl_depth %d, total_len %u, " 399 pr_debug("%s - page shift %d, pbl_depth %d, total_len %u, "
398 "*iova_start %llx, first pa %llx, last pa %llx\n", 400 "*iova_start %llx, first pa %llx, last pa %llx\n",
399 __FUNCTION__, page_shift, pbl_depth, total_len, 401 __FUNCTION__, page_shift, pbl_depth, total_len,
400 *iova_start, page_list[0], page_list[pbl_depth-1]); 402 (unsigned long long) *iova_start,
403 (unsigned long long) page_list[0],
404 (unsigned long long) page_list[pbl_depth-1]);
401 err = c2_nsmr_register_phys_kern(to_c2dev(ib_pd->device), page_list, 405 err = c2_nsmr_register_phys_kern(to_c2dev(ib_pd->device), page_list,
402 (1 << page_shift), pbl_depth, 406 (1 << page_shift), pbl_depth,
403 total_len, 0, iova_start, 407 total_len, 0, iova_start,
diff --git a/drivers/infiniband/hw/amso1100/c2_rnic.c b/drivers/infiniband/hw/amso1100/c2_rnic.c
index f49a32b7a8f6..e37c5688c214 100644
--- a/drivers/infiniband/hw/amso1100/c2_rnic.c
+++ b/drivers/infiniband/hw/amso1100/c2_rnic.c
@@ -527,7 +527,7 @@ int c2_rnic_init(struct c2_dev *c2dev)
527 DMA_FROM_DEVICE); 527 DMA_FROM_DEVICE);
528 pci_unmap_addr_set(&c2dev->rep_vq, mapping, c2dev->rep_vq.host_dma); 528 pci_unmap_addr_set(&c2dev->rep_vq, mapping, c2dev->rep_vq.host_dma);
529 pr_debug("%s rep_vq va %p dma %llx\n", __FUNCTION__, q1_pages, 529 pr_debug("%s rep_vq va %p dma %llx\n", __FUNCTION__, q1_pages,
530 (u64)c2dev->rep_vq.host_dma); 530 (unsigned long long) c2dev->rep_vq.host_dma);
531 c2_mq_rep_init(&c2dev->rep_vq, 531 c2_mq_rep_init(&c2dev->rep_vq,
532 1, 532 1,
533 qsize, 533 qsize,
@@ -550,7 +550,7 @@ int c2_rnic_init(struct c2_dev *c2dev)
550 DMA_FROM_DEVICE); 550 DMA_FROM_DEVICE);
551 pci_unmap_addr_set(&c2dev->aeq, mapping, c2dev->aeq.host_dma); 551 pci_unmap_addr_set(&c2dev->aeq, mapping, c2dev->aeq.host_dma);
552 pr_debug("%s aeq va %p dma %llx\n", __FUNCTION__, q1_pages, 552 pr_debug("%s aeq va %p dma %llx\n", __FUNCTION__, q1_pages,
553 (u64)c2dev->rep_vq.host_dma); 553 (unsigned long long) c2dev->rep_vq.host_dma);
554 c2_mq_rep_init(&c2dev->aeq, 554 c2_mq_rep_init(&c2dev->aeq,
555 2, 555 2,
556 qsize, 556 qsize,
diff --git a/drivers/infiniband/hw/ipath/ipath_common.h b/drivers/infiniband/hw/ipath/ipath_common.h
index f577905e3aca..54139d398181 100644
--- a/drivers/infiniband/hw/ipath/ipath_common.h
+++ b/drivers/infiniband/hw/ipath/ipath_common.h
@@ -141,8 +141,9 @@ struct infinipath_stats {
141 * packets if ipath not configured, etc.) 141 * packets if ipath not configured, etc.)
142 */ 142 */
143 __u64 sps_krdrops; 143 __u64 sps_krdrops;
144 __u64 sps_txeparity; /* PIO buffer parity error, recovered */
144 /* pad for future growth */ 145 /* pad for future growth */
145 __u64 __sps_pad[46]; 146 __u64 __sps_pad[45];
146}; 147};
147 148
148/* 149/*
@@ -185,6 +186,9 @@ typedef enum _ipath_ureg {
185#define IPATH_RUNTIME_PCIE 0x2 186#define IPATH_RUNTIME_PCIE 0x2
186#define IPATH_RUNTIME_FORCE_WC_ORDER 0x4 187#define IPATH_RUNTIME_FORCE_WC_ORDER 0x4
187#define IPATH_RUNTIME_RCVHDR_COPY 0x8 188#define IPATH_RUNTIME_RCVHDR_COPY 0x8
189#define IPATH_RUNTIME_MASTER 0x10
190#define IPATH_RUNTIME_PBC_REWRITE 0x20
191#define IPATH_RUNTIME_LOOSE_DMA_ALIGN 0x40
188 192
189/* 193/*
190 * This structure is returned by ipath_userinit() immediately after 194 * This structure is returned by ipath_userinit() immediately after
@@ -202,7 +206,8 @@ struct ipath_base_info {
202 /* version of software, for feature checking. */ 206 /* version of software, for feature checking. */
203 __u32 spi_sw_version; 207 __u32 spi_sw_version;
204 /* InfiniPath port assigned, goes into sent packets */ 208 /* InfiniPath port assigned, goes into sent packets */
205 __u32 spi_port; 209 __u16 spi_port;
210 __u16 spi_subport;
206 /* 211 /*
207 * IB MTU, packets IB data must be less than this. 212 * IB MTU, packets IB data must be less than this.
208 * The MTU is in bytes, and will be a multiple of 4 bytes. 213 * The MTU is in bytes, and will be a multiple of 4 bytes.
@@ -218,7 +223,7 @@ struct ipath_base_info {
218 __u32 spi_tidcnt; 223 __u32 spi_tidcnt;
219 /* size of the TID Eager list in infinipath, in entries */ 224 /* size of the TID Eager list in infinipath, in entries */
220 __u32 spi_tidegrcnt; 225 __u32 spi_tidegrcnt;
221 /* size of a single receive header queue entry. */ 226 /* size of a single receive header queue entry in words. */
222 __u32 spi_rcvhdrent_size; 227 __u32 spi_rcvhdrent_size;
223 /* 228 /*
224 * Count of receive header queue entries allocated. 229 * Count of receive header queue entries allocated.
@@ -310,6 +315,12 @@ struct ipath_base_info {
310 __u32 spi_filler_for_align; 315 __u32 spi_filler_for_align;
311 /* address of readonly memory copy of the rcvhdrq tail register. */ 316 /* address of readonly memory copy of the rcvhdrq tail register. */
312 __u64 spi_rcvhdr_tailaddr; 317 __u64 spi_rcvhdr_tailaddr;
318
319 /* shared memory pages for subports if IPATH_RUNTIME_MASTER is set */
320 __u64 spi_subport_uregbase;
321 __u64 spi_subport_rcvegrbuf;
322 __u64 spi_subport_rcvhdr_base;
323
313} __attribute__ ((aligned(8))); 324} __attribute__ ((aligned(8)));
314 325
315 326
@@ -328,12 +339,12 @@ struct ipath_base_info {
328 339
329/* 340/*
330 * Minor version differences are always compatible 341 * Minor version differences are always compatible
331 * a within a major version, however if if user software is larger 342 * a within a major version, however if user software is larger
332 * than driver software, some new features and/or structure fields 343 * than driver software, some new features and/or structure fields
333 * may not be implemented; the user code must deal with this if it 344 * may not be implemented; the user code must deal with this if it
334 * cares, or it must abort after initialization reports the difference 345 * cares, or it must abort after initialization reports the difference.
335 */ 346 */
336#define IPATH_USER_SWMINOR 2 347#define IPATH_USER_SWMINOR 3
337 348
338#define IPATH_USER_SWVERSION ((IPATH_USER_SWMAJOR<<16) | IPATH_USER_SWMINOR) 349#define IPATH_USER_SWVERSION ((IPATH_USER_SWMAJOR<<16) | IPATH_USER_SWMINOR)
339 350
@@ -379,7 +390,16 @@ struct ipath_user_info {
379 */ 390 */
380 __u32 spu_rcvhdrsize; 391 __u32 spu_rcvhdrsize;
381 392
382 __u64 spu_unused; /* kept for compatible layout */ 393 /*
394 * If two or more processes wish to share a port, each process
395 * must set the spu_subport_cnt and spu_subport_id to the same
396 * values. The only restriction on the spu_subport_id is that
397 * it be unique for a given node.
398 */
399 __u16 spu_subport_cnt;
400 __u16 spu_subport_id;
401
402 __u32 spu_unused; /* kept for compatible layout */
383 403
384 /* 404 /*
385 * address of struct base_info to write to 405 * address of struct base_info to write to
@@ -392,19 +412,25 @@ struct ipath_user_info {
392 412
393#define IPATH_CMD_MIN 16 413#define IPATH_CMD_MIN 16
394 414
395#define IPATH_CMD_USER_INIT 16 /* set up userspace */ 415#define __IPATH_CMD_USER_INIT 16 /* old set up userspace (for old user code) */
396#define IPATH_CMD_PORT_INFO 17 /* find out what resources we got */ 416#define IPATH_CMD_PORT_INFO 17 /* find out what resources we got */
397#define IPATH_CMD_RECV_CTRL 18 /* control receipt of packets */ 417#define IPATH_CMD_RECV_CTRL 18 /* control receipt of packets */
398#define IPATH_CMD_TID_UPDATE 19 /* update expected TID entries */ 418#define IPATH_CMD_TID_UPDATE 19 /* update expected TID entries */
399#define IPATH_CMD_TID_FREE 20 /* free expected TID entries */ 419#define IPATH_CMD_TID_FREE 20 /* free expected TID entries */
400#define IPATH_CMD_SET_PART_KEY 21 /* add partition key */ 420#define IPATH_CMD_SET_PART_KEY 21 /* add partition key */
421#define IPATH_CMD_SLAVE_INFO 22 /* return info on slave processes */
422#define IPATH_CMD_ASSIGN_PORT 23 /* allocate HCA and port */
423#define IPATH_CMD_USER_INIT 24 /* set up userspace */
401 424
402#define IPATH_CMD_MAX 21 425#define IPATH_CMD_MAX 24
403 426
404struct ipath_port_info { 427struct ipath_port_info {
405 __u32 num_active; /* number of active units */ 428 __u32 num_active; /* number of active units */
406 __u32 unit; /* unit (chip) assigned to caller */ 429 __u32 unit; /* unit (chip) assigned to caller */
407 __u32 port; /* port on unit assigned to caller */ 430 __u16 port; /* port on unit assigned to caller */
431 __u16 subport; /* subport on unit assigned to caller */
432 __u16 num_ports; /* number of ports available on unit */
433 __u16 num_subports; /* number of subport slaves opened on port */
408}; 434};
409 435
410struct ipath_tid_info { 436struct ipath_tid_info {
@@ -435,6 +461,8 @@ struct ipath_cmd {
435 __u32 recv_ctrl; 461 __u32 recv_ctrl;
436 /* partition key to set */ 462 /* partition key to set */
437 __u16 part_key; 463 __u16 part_key;
464 /* user address of __u32 bitmask of active slaves */
465 __u64 slave_mask_addr;
438 } cmd; 466 } cmd;
439}; 467};
440 468
@@ -596,6 +624,10 @@ struct infinipath_counters {
596 624
597/* K_PktFlags bits */ 625/* K_PktFlags bits */
598#define INFINIPATH_KPF_INTR 0x1 626#define INFINIPATH_KPF_INTR 0x1
627#define INFINIPATH_KPF_SUBPORT_MASK 0x3
628#define INFINIPATH_KPF_SUBPORT_SHIFT 1
629
630#define INFINIPATH_MAX_SUBPORT 4
599 631
600/* SendPIO per-buffer control */ 632/* SendPIO per-buffer control */
601#define INFINIPATH_SP_TEST 0x40 633#define INFINIPATH_SP_TEST 0x40
@@ -610,7 +642,7 @@ struct ipath_header {
610 /* 642 /*
611 * Version - 4 bits, Port - 4 bits, TID - 10 bits and Offset - 643 * Version - 4 bits, Port - 4 bits, TID - 10 bits and Offset -
612 * 14 bits before ECO change ~28 Dec 03. After that, Vers 4, 644 * 14 bits before ECO change ~28 Dec 03. After that, Vers 4,
613 * Port 3, TID 11, offset 14. 645 * Port 4, TID 11, offset 13.
614 */ 646 */
615 __le32 ver_port_tid_offset; 647 __le32 ver_port_tid_offset;
616 __le16 chksum; 648 __le16 chksum;
diff --git a/drivers/infiniband/hw/ipath/ipath_cq.c b/drivers/infiniband/hw/ipath/ipath_cq.c
index 049221bc590e..87462e0cb4d2 100644
--- a/drivers/infiniband/hw/ipath/ipath_cq.c
+++ b/drivers/infiniband/hw/ipath/ipath_cq.c
@@ -46,7 +46,7 @@
46 */ 46 */
47void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int solicited) 47void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int solicited)
48{ 48{
49 struct ipath_cq_wc *wc = cq->queue; 49 struct ipath_cq_wc *wc;
50 unsigned long flags; 50 unsigned long flags;
51 u32 head; 51 u32 head;
52 u32 next; 52 u32 next;
@@ -57,6 +57,7 @@ void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int solicited)
57 * Note that the head pointer might be writable by user processes. 57 * Note that the head pointer might be writable by user processes.
58 * Take care to verify it is a sane value. 58 * Take care to verify it is a sane value.
59 */ 59 */
60 wc = cq->queue;
60 head = wc->head; 61 head = wc->head;
61 if (head >= (unsigned) cq->ibcq.cqe) { 62 if (head >= (unsigned) cq->ibcq.cqe) {
62 head = cq->ibcq.cqe; 63 head = cq->ibcq.cqe;
@@ -109,21 +110,27 @@ void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int solicited)
109int ipath_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry) 110int ipath_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
110{ 111{
111 struct ipath_cq *cq = to_icq(ibcq); 112 struct ipath_cq *cq = to_icq(ibcq);
112 struct ipath_cq_wc *wc = cq->queue; 113 struct ipath_cq_wc *wc;
113 unsigned long flags; 114 unsigned long flags;
114 int npolled; 115 int npolled;
116 u32 tail;
115 117
116 spin_lock_irqsave(&cq->lock, flags); 118 spin_lock_irqsave(&cq->lock, flags);
117 119
120 wc = cq->queue;
121 tail = wc->tail;
122 if (tail > (u32) cq->ibcq.cqe)
123 tail = (u32) cq->ibcq.cqe;
118 for (npolled = 0; npolled < num_entries; ++npolled, ++entry) { 124 for (npolled = 0; npolled < num_entries; ++npolled, ++entry) {
119 if (wc->tail == wc->head) 125 if (tail == wc->head)
120 break; 126 break;
121 *entry = wc->queue[wc->tail]; 127 *entry = wc->queue[tail];
122 if (wc->tail >= cq->ibcq.cqe) 128 if (tail >= cq->ibcq.cqe)
123 wc->tail = 0; 129 tail = 0;
124 else 130 else
125 wc->tail++; 131 tail++;
126 } 132 }
133 wc->tail = tail;
127 134
128 spin_unlock_irqrestore(&cq->lock, flags); 135 spin_unlock_irqrestore(&cq->lock, flags);
129 136
@@ -177,11 +184,6 @@ struct ib_cq *ipath_create_cq(struct ib_device *ibdev, int entries,
177 goto done; 184 goto done;
178 } 185 }
179 186
180 if (dev->n_cqs_allocated == ib_ipath_max_cqs) {
181 ret = ERR_PTR(-ENOMEM);
182 goto done;
183 }
184
185 /* Allocate the completion queue structure. */ 187 /* Allocate the completion queue structure. */
186 cq = kmalloc(sizeof(*cq), GFP_KERNEL); 188 cq = kmalloc(sizeof(*cq), GFP_KERNEL);
187 if (!cq) { 189 if (!cq) {
@@ -237,6 +239,16 @@ struct ib_cq *ipath_create_cq(struct ib_device *ibdev, int entries,
237 } else 239 } else
238 cq->ip = NULL; 240 cq->ip = NULL;
239 241
242 spin_lock(&dev->n_cqs_lock);
243 if (dev->n_cqs_allocated == ib_ipath_max_cqs) {
244 spin_unlock(&dev->n_cqs_lock);
245 ret = ERR_PTR(-ENOMEM);
246 goto bail_wc;
247 }
248
249 dev->n_cqs_allocated++;
250 spin_unlock(&dev->n_cqs_lock);
251
240 /* 252 /*
241 * ib_create_cq() will initialize cq->ibcq except for cq->ibcq.cqe. 253 * ib_create_cq() will initialize cq->ibcq except for cq->ibcq.cqe.
242 * The number of entries should be >= the number requested or return 254 * The number of entries should be >= the number requested or return
@@ -253,7 +265,6 @@ struct ib_cq *ipath_create_cq(struct ib_device *ibdev, int entries,
253 265
254 ret = &cq->ibcq; 266 ret = &cq->ibcq;
255 267
256 dev->n_cqs_allocated++;
257 goto done; 268 goto done;
258 269
259bail_wc: 270bail_wc:
@@ -280,7 +291,9 @@ int ipath_destroy_cq(struct ib_cq *ibcq)
280 struct ipath_cq *cq = to_icq(ibcq); 291 struct ipath_cq *cq = to_icq(ibcq);
281 292
282 tasklet_kill(&cq->comptask); 293 tasklet_kill(&cq->comptask);
294 spin_lock(&dev->n_cqs_lock);
283 dev->n_cqs_allocated--; 295 dev->n_cqs_allocated--;
296 spin_unlock(&dev->n_cqs_lock);
284 if (cq->ip) 297 if (cq->ip)
285 kref_put(&cq->ip->ref, ipath_release_mmap_info); 298 kref_put(&cq->ip->ref, ipath_release_mmap_info);
286 else 299 else
@@ -316,10 +329,16 @@ int ipath_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify notify)
316 return 0; 329 return 0;
317} 330}
318 331
332/**
333 * ipath_resize_cq - change the size of the CQ
334 * @ibcq: the completion queue
335 *
336 * Returns 0 for success.
337 */
319int ipath_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata) 338int ipath_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata)
320{ 339{
321 struct ipath_cq *cq = to_icq(ibcq); 340 struct ipath_cq *cq = to_icq(ibcq);
322 struct ipath_cq_wc *old_wc = cq->queue; 341 struct ipath_cq_wc *old_wc;
323 struct ipath_cq_wc *wc; 342 struct ipath_cq_wc *wc;
324 u32 head, tail, n; 343 u32 head, tail, n;
325 int ret; 344 int ret;
@@ -355,6 +374,7 @@ int ipath_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata)
355 * Make sure head and tail are sane since they 374 * Make sure head and tail are sane since they
356 * might be user writable. 375 * might be user writable.
357 */ 376 */
377 old_wc = cq->queue;
358 head = old_wc->head; 378 head = old_wc->head;
359 if (head > (u32) cq->ibcq.cqe) 379 if (head > (u32) cq->ibcq.cqe)
360 head = (u32) cq->ibcq.cqe; 380 head = (u32) cq->ibcq.cqe;
diff --git a/drivers/infiniband/hw/ipath/ipath_driver.c b/drivers/infiniband/hw/ipath/ipath_driver.c
index 2108466c7e33..12cefa658f3b 100644
--- a/drivers/infiniband/hw/ipath/ipath_driver.c
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c
@@ -95,16 +95,6 @@ const char *ipath_ibcstatus_str[] = {
95 "RecovIdle", 95 "RecovIdle",
96}; 96};
97 97
98/*
99 * These variables are initialized in the chip-specific files
100 * but are defined here.
101 */
102u16 ipath_gpio_sda_num, ipath_gpio_scl_num;
103u64 ipath_gpio_sda, ipath_gpio_scl;
104u64 infinipath_i_bitsextant;
105ipath_err_t infinipath_e_bitsextant, infinipath_hwe_bitsextant;
106u32 infinipath_i_rcvavail_mask, infinipath_i_rcvurg_mask;
107
108static void __devexit ipath_remove_one(struct pci_dev *); 98static void __devexit ipath_remove_one(struct pci_dev *);
109static int __devinit ipath_init_one(struct pci_dev *, 99static int __devinit ipath_init_one(struct pci_dev *,
110 const struct pci_device_id *); 100 const struct pci_device_id *);
@@ -527,28 +517,146 @@ bail:
527 return ret; 517 return ret;
528} 518}
529 519
520static void __devexit cleanup_device(struct ipath_devdata *dd)
521{
522 int port;
523
524 ipath_shutdown_device(dd);
525
526 if (*dd->ipath_statusp & IPATH_STATUS_CHIP_PRESENT) {
527 /* can't do anything more with chip; needs re-init */
528 *dd->ipath_statusp &= ~IPATH_STATUS_CHIP_PRESENT;
529 if (dd->ipath_kregbase) {
530 /*
531 * if we haven't already cleaned up before these are
532 * to ensure any register reads/writes "fail" until
533 * re-init
534 */
535 dd->ipath_kregbase = NULL;
536 dd->ipath_uregbase = 0;
537 dd->ipath_sregbase = 0;
538 dd->ipath_cregbase = 0;
539 dd->ipath_kregsize = 0;
540 }
541 ipath_disable_wc(dd);
542 }
543
544 if (dd->ipath_pioavailregs_dma) {
545 dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
546 (void *) dd->ipath_pioavailregs_dma,
547 dd->ipath_pioavailregs_phys);
548 dd->ipath_pioavailregs_dma = NULL;
549 }
550 if (dd->ipath_dummy_hdrq) {
551 dma_free_coherent(&dd->pcidev->dev,
552 dd->ipath_pd[0]->port_rcvhdrq_size,
553 dd->ipath_dummy_hdrq, dd->ipath_dummy_hdrq_phys);
554 dd->ipath_dummy_hdrq = NULL;
555 }
556
557 if (dd->ipath_pageshadow) {
558 struct page **tmpp = dd->ipath_pageshadow;
559 dma_addr_t *tmpd = dd->ipath_physshadow;
560 int i, cnt = 0;
561
562 ipath_cdbg(VERBOSE, "Unlocking any expTID pages still "
563 "locked\n");
564 for (port = 0; port < dd->ipath_cfgports; port++) {
565 int port_tidbase = port * dd->ipath_rcvtidcnt;
566 int maxtid = port_tidbase + dd->ipath_rcvtidcnt;
567 for (i = port_tidbase; i < maxtid; i++) {
568 if (!tmpp[i])
569 continue;
570 pci_unmap_page(dd->pcidev, tmpd[i],
571 PAGE_SIZE, PCI_DMA_FROMDEVICE);
572 ipath_release_user_pages(&tmpp[i], 1);
573 tmpp[i] = NULL;
574 cnt++;
575 }
576 }
577 if (cnt) {
578 ipath_stats.sps_pageunlocks += cnt;
579 ipath_cdbg(VERBOSE, "There were still %u expTID "
580 "entries locked\n", cnt);
581 }
582 if (ipath_stats.sps_pagelocks ||
583 ipath_stats.sps_pageunlocks)
584 ipath_cdbg(VERBOSE, "%llu pages locked, %llu "
585 "unlocked via ipath_m{un}lock\n",
586 (unsigned long long)
587 ipath_stats.sps_pagelocks,
588 (unsigned long long)
589 ipath_stats.sps_pageunlocks);
590
591 ipath_cdbg(VERBOSE, "Free shadow page tid array at %p\n",
592 dd->ipath_pageshadow);
593 vfree(dd->ipath_pageshadow);
594 dd->ipath_pageshadow = NULL;
595 }
596
597 /*
598 * free any resources still in use (usually just kernel ports)
599 * at unload; we do for portcnt, not cfgports, because cfgports
600 * could have changed while we were loaded.
601 */
602 for (port = 0; port < dd->ipath_portcnt; port++) {
603 struct ipath_portdata *pd = dd->ipath_pd[port];
604 dd->ipath_pd[port] = NULL;
605 ipath_free_pddata(dd, pd);
606 }
607 kfree(dd->ipath_pd);
608 /*
609 * debuggability, in case some cleanup path tries to use it
610 * after this
611 */
612 dd->ipath_pd = NULL;
613}
614
530static void __devexit ipath_remove_one(struct pci_dev *pdev) 615static void __devexit ipath_remove_one(struct pci_dev *pdev)
531{ 616{
532 struct ipath_devdata *dd; 617 struct ipath_devdata *dd = pci_get_drvdata(pdev);
533 618
534 ipath_cdbg(VERBOSE, "removing, pdev=%p\n", pdev); 619 ipath_cdbg(VERBOSE, "removing, pdev=%p, dd=%p\n", pdev, dd);
535 if (!pdev) 620
536 return; 621 if (dd->verbs_dev)
622 ipath_unregister_ib_device(dd->verbs_dev);
537 623
538 dd = pci_get_drvdata(pdev);
539 ipath_unregister_ib_device(dd->verbs_dev);
540 ipath_diag_remove(dd); 624 ipath_diag_remove(dd);
541 ipath_user_remove(dd); 625 ipath_user_remove(dd);
542 ipathfs_remove_device(dd); 626 ipathfs_remove_device(dd);
543 ipath_device_remove_group(&pdev->dev, dd); 627 ipath_device_remove_group(&pdev->dev, dd);
628
544 ipath_cdbg(VERBOSE, "Releasing pci memory regions, dd %p, " 629 ipath_cdbg(VERBOSE, "Releasing pci memory regions, dd %p, "
545 "unit %u\n", dd, (u32) dd->ipath_unit); 630 "unit %u\n", dd, (u32) dd->ipath_unit);
546 if (dd->ipath_kregbase) { 631
547 ipath_cdbg(VERBOSE, "Unmapping kregbase %p\n", 632 cleanup_device(dd);
548 dd->ipath_kregbase); 633
549 iounmap((volatile void __iomem *) dd->ipath_kregbase); 634 /*
550 dd->ipath_kregbase = NULL; 635 * turn off rcv, send, and interrupts for all ports, all drivers
551 } 636 * should also hard reset the chip here?
637 * free up port 0 (kernel) rcvhdr, egr bufs, and eventually tid bufs
638 * for all versions of the driver, if they were allocated
639 */
640 if (pdev->irq) {
641 ipath_cdbg(VERBOSE,
642 "unit %u free_irq of irq %x\n",
643 dd->ipath_unit, pdev->irq);
644 free_irq(pdev->irq, dd);
645 } else
646 ipath_dbg("irq is 0, not doing free_irq "
647 "for unit %u\n", dd->ipath_unit);
648 /*
649 * we check for NULL here, because it's outside
650 * the kregbase check, and we need to call it
651 * after the free_irq. Thus it's possible that
652 * the function pointers were never initialized.
653 */
654 if (dd->ipath_f_cleanup)
655 /* clean up chip-specific stuff */
656 dd->ipath_f_cleanup(dd);
657
658 ipath_cdbg(VERBOSE, "Unmapping kregbase %p\n", dd->ipath_kregbase);
659 iounmap((volatile void __iomem *) dd->ipath_kregbase);
552 pci_release_regions(pdev); 660 pci_release_regions(pdev);
553 ipath_cdbg(VERBOSE, "calling pci_disable_device\n"); 661 ipath_cdbg(VERBOSE, "calling pci_disable_device\n");
554 pci_disable_device(pdev); 662 pci_disable_device(pdev);
@@ -760,8 +868,8 @@ static void get_rhf_errstring(u32 err, char *msg, size_t len)
760static inline void *ipath_get_egrbuf(struct ipath_devdata *dd, u32 bufnum, 868static inline void *ipath_get_egrbuf(struct ipath_devdata *dd, u32 bufnum,
761 int err) 869 int err)
762{ 870{
763 return dd->ipath_port0_skbs ? 871 return dd->ipath_port0_skbinfo ?
764 (void *)dd->ipath_port0_skbs[bufnum]->data : NULL; 872 (void *) dd->ipath_port0_skbinfo[bufnum].skb->data : NULL;
765} 873}
766 874
767/** 875/**
@@ -783,31 +891,34 @@ struct sk_buff *ipath_alloc_skb(struct ipath_devdata *dd,
783 */ 891 */
784 892
785 /* 893 /*
786 * We need 4 extra bytes for unaligned transfer copying 894 * We need 2 extra bytes for ipath_ether data sent in the
895 * key header. In order to keep everything dword aligned,
896 * we'll reserve 4 bytes.
787 */ 897 */
898 len = dd->ipath_ibmaxlen + 4;
899
788 if (dd->ipath_flags & IPATH_4BYTE_TID) { 900 if (dd->ipath_flags & IPATH_4BYTE_TID) {
789 /* we need a 4KB multiple alignment, and there is no way 901 /* We need a 2KB multiple alignment, and there is no way
790 * to do it except to allocate extra and then skb_reserve 902 * to do it except to allocate extra and then skb_reserve
791 * enough to bring it up to the right alignment. 903 * enough to bring it up to the right alignment.
792 */ 904 */
793 len = dd->ipath_ibmaxlen + 4 + (1 << 11) - 1; 905 len += 2047;
794 } 906 }
795 else 907
796 len = dd->ipath_ibmaxlen + 4;
797 skb = __dev_alloc_skb(len, gfp_mask); 908 skb = __dev_alloc_skb(len, gfp_mask);
798 if (!skb) { 909 if (!skb) {
799 ipath_dev_err(dd, "Failed to allocate skbuff, length %u\n", 910 ipath_dev_err(dd, "Failed to allocate skbuff, length %u\n",
800 len); 911 len);
801 goto bail; 912 goto bail;
802 } 913 }
914
915 skb_reserve(skb, 4);
916
803 if (dd->ipath_flags & IPATH_4BYTE_TID) { 917 if (dd->ipath_flags & IPATH_4BYTE_TID) {
804 u32 una = ((1 << 11) - 1) & (unsigned long)(skb->data + 4); 918 u32 una = (unsigned long)skb->data & 2047;
805 if (una) 919 if (una)
806 skb_reserve(skb, 4 + (1 << 11) - una); 920 skb_reserve(skb, 2048 - una);
807 else 921 }
808 skb_reserve(skb, 4);
809 } else
810 skb_reserve(skb, 4);
811 922
812bail: 923bail:
813 return skb; 924 return skb;
@@ -1326,6 +1437,9 @@ int ipath_create_rcvhdrq(struct ipath_devdata *dd,
1326 "for port %u rcvhdrqtailaddr failed\n", 1437 "for port %u rcvhdrqtailaddr failed\n",
1327 pd->port_port); 1438 pd->port_port);
1328 ret = -ENOMEM; 1439 ret = -ENOMEM;
1440 dma_free_coherent(&dd->pcidev->dev, amt,
1441 pd->port_rcvhdrq, pd->port_rcvhdrq_phys);
1442 pd->port_rcvhdrq = NULL;
1329 goto bail; 1443 goto bail;
1330 } 1444 }
1331 pd->port_rcvhdrqtailaddr_phys = phys_hdrqtail; 1445 pd->port_rcvhdrqtailaddr_phys = phys_hdrqtail;
@@ -1347,12 +1461,13 @@ int ipath_create_rcvhdrq(struct ipath_devdata *dd,
1347 ipath_cdbg(VERBOSE, "reuse port %d rcvhdrq @%p %llx phys; " 1461 ipath_cdbg(VERBOSE, "reuse port %d rcvhdrq @%p %llx phys; "
1348 "hdrtailaddr@%p %llx physical\n", 1462 "hdrtailaddr@%p %llx physical\n",
1349 pd->port_port, pd->port_rcvhdrq, 1463 pd->port_port, pd->port_rcvhdrq,
1350 pd->port_rcvhdrq_phys, pd->port_rcvhdrtail_kvaddr, 1464 (unsigned long long) pd->port_rcvhdrq_phys,
1351 (unsigned long long)pd->port_rcvhdrqtailaddr_phys); 1465 pd->port_rcvhdrtail_kvaddr, (unsigned long long)
1466 pd->port_rcvhdrqtailaddr_phys);
1352 1467
1353 /* clear for security and sanity on each use */ 1468 /* clear for security and sanity on each use */
1354 memset(pd->port_rcvhdrq, 0, pd->port_rcvhdrq_size); 1469 memset(pd->port_rcvhdrq, 0, pd->port_rcvhdrq_size);
1355 memset((void *)pd->port_rcvhdrtail_kvaddr, 0, PAGE_SIZE); 1470 memset(pd->port_rcvhdrtail_kvaddr, 0, PAGE_SIZE);
1356 1471
1357 /* 1472 /*
1358 * tell chip each time we init it, even if we are re-using previous 1473 * tell chip each time we init it, even if we are re-using previous
@@ -1805,7 +1920,7 @@ void ipath_free_pddata(struct ipath_devdata *dd, struct ipath_portdata *pd)
1805 pd->port_rcvhdrq = NULL; 1920 pd->port_rcvhdrq = NULL;
1806 if (pd->port_rcvhdrtail_kvaddr) { 1921 if (pd->port_rcvhdrtail_kvaddr) {
1807 dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE, 1922 dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
1808 (void *)pd->port_rcvhdrtail_kvaddr, 1923 pd->port_rcvhdrtail_kvaddr,
1809 pd->port_rcvhdrqtailaddr_phys); 1924 pd->port_rcvhdrqtailaddr_phys);
1810 pd->port_rcvhdrtail_kvaddr = NULL; 1925 pd->port_rcvhdrtail_kvaddr = NULL;
1811 } 1926 }
@@ -1824,24 +1939,32 @@ void ipath_free_pddata(struct ipath_devdata *dd, struct ipath_portdata *pd)
1824 dma_free_coherent(&dd->pcidev->dev, size, 1939 dma_free_coherent(&dd->pcidev->dev, size,
1825 base, pd->port_rcvegrbuf_phys[e]); 1940 base, pd->port_rcvegrbuf_phys[e]);
1826 } 1941 }
1827 vfree(pd->port_rcvegrbuf); 1942 kfree(pd->port_rcvegrbuf);
1828 pd->port_rcvegrbuf = NULL; 1943 pd->port_rcvegrbuf = NULL;
1829 vfree(pd->port_rcvegrbuf_phys); 1944 kfree(pd->port_rcvegrbuf_phys);
1830 pd->port_rcvegrbuf_phys = NULL; 1945 pd->port_rcvegrbuf_phys = NULL;
1831 pd->port_rcvegrbuf_chunks = 0; 1946 pd->port_rcvegrbuf_chunks = 0;
1832 } else if (pd->port_port == 0 && dd->ipath_port0_skbs) { 1947 } else if (pd->port_port == 0 && dd->ipath_port0_skbinfo) {
1833 unsigned e; 1948 unsigned e;
1834 struct sk_buff **skbs = dd->ipath_port0_skbs; 1949 struct ipath_skbinfo *skbinfo = dd->ipath_port0_skbinfo;
1835 1950
1836 dd->ipath_port0_skbs = NULL; 1951 dd->ipath_port0_skbinfo = NULL;
1837 ipath_cdbg(VERBOSE, "free closed port %d ipath_port0_skbs " 1952 ipath_cdbg(VERBOSE, "free closed port %d "
1838 "@ %p\n", pd->port_port, skbs); 1953 "ipath_port0_skbinfo @ %p\n", pd->port_port,
1954 skbinfo);
1839 for (e = 0; e < dd->ipath_rcvegrcnt; e++) 1955 for (e = 0; e < dd->ipath_rcvegrcnt; e++)
1840 if (skbs[e]) 1956 if (skbinfo[e].skb) {
1841 dev_kfree_skb(skbs[e]); 1957 pci_unmap_single(dd->pcidev, skbinfo[e].phys,
1842 vfree(skbs); 1958 dd->ipath_ibmaxlen,
1959 PCI_DMA_FROMDEVICE);
1960 dev_kfree_skb(skbinfo[e].skb);
1961 }
1962 vfree(skbinfo);
1843 } 1963 }
1844 kfree(pd->port_tid_pg_list); 1964 kfree(pd->port_tid_pg_list);
1965 vfree(pd->subport_uregbase);
1966 vfree(pd->subport_rcvegrbuf);
1967 vfree(pd->subport_rcvhdr_base);
1845 kfree(pd); 1968 kfree(pd);
1846} 1969}
1847 1970
@@ -1907,150 +2030,12 @@ bail:
1907 return ret; 2030 return ret;
1908} 2031}
1909 2032
1910static void cleanup_device(struct ipath_devdata *dd)
1911{
1912 int port;
1913
1914 ipath_shutdown_device(dd);
1915
1916 if (*dd->ipath_statusp & IPATH_STATUS_CHIP_PRESENT) {
1917 /* can't do anything more with chip; needs re-init */
1918 *dd->ipath_statusp &= ~IPATH_STATUS_CHIP_PRESENT;
1919 if (dd->ipath_kregbase) {
1920 /*
1921 * if we haven't already cleaned up before these are
1922 * to ensure any register reads/writes "fail" until
1923 * re-init
1924 */
1925 dd->ipath_kregbase = NULL;
1926 dd->ipath_uregbase = 0;
1927 dd->ipath_sregbase = 0;
1928 dd->ipath_cregbase = 0;
1929 dd->ipath_kregsize = 0;
1930 }
1931 ipath_disable_wc(dd);
1932 }
1933
1934 if (dd->ipath_pioavailregs_dma) {
1935 dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
1936 (void *) dd->ipath_pioavailregs_dma,
1937 dd->ipath_pioavailregs_phys);
1938 dd->ipath_pioavailregs_dma = NULL;
1939 }
1940 if (dd->ipath_dummy_hdrq) {
1941 dma_free_coherent(&dd->pcidev->dev,
1942 dd->ipath_pd[0]->port_rcvhdrq_size,
1943 dd->ipath_dummy_hdrq, dd->ipath_dummy_hdrq_phys);
1944 dd->ipath_dummy_hdrq = NULL;
1945 }
1946
1947 if (dd->ipath_pageshadow) {
1948 struct page **tmpp = dd->ipath_pageshadow;
1949 int i, cnt = 0;
1950
1951 ipath_cdbg(VERBOSE, "Unlocking any expTID pages still "
1952 "locked\n");
1953 for (port = 0; port < dd->ipath_cfgports; port++) {
1954 int port_tidbase = port * dd->ipath_rcvtidcnt;
1955 int maxtid = port_tidbase + dd->ipath_rcvtidcnt;
1956 for (i = port_tidbase; i < maxtid; i++) {
1957 if (!tmpp[i])
1958 continue;
1959 ipath_release_user_pages(&tmpp[i], 1);
1960 tmpp[i] = NULL;
1961 cnt++;
1962 }
1963 }
1964 if (cnt) {
1965 ipath_stats.sps_pageunlocks += cnt;
1966 ipath_cdbg(VERBOSE, "There were still %u expTID "
1967 "entries locked\n", cnt);
1968 }
1969 if (ipath_stats.sps_pagelocks ||
1970 ipath_stats.sps_pageunlocks)
1971 ipath_cdbg(VERBOSE, "%llu pages locked, %llu "
1972 "unlocked via ipath_m{un}lock\n",
1973 (unsigned long long)
1974 ipath_stats.sps_pagelocks,
1975 (unsigned long long)
1976 ipath_stats.sps_pageunlocks);
1977
1978 ipath_cdbg(VERBOSE, "Free shadow page tid array at %p\n",
1979 dd->ipath_pageshadow);
1980 vfree(dd->ipath_pageshadow);
1981 dd->ipath_pageshadow = NULL;
1982 }
1983
1984 /*
1985 * free any resources still in use (usually just kernel ports)
1986 * at unload; we do for portcnt, not cfgports, because cfgports
1987 * could have changed while we were loaded.
1988 */
1989 for (port = 0; port < dd->ipath_portcnt; port++) {
1990 struct ipath_portdata *pd = dd->ipath_pd[port];
1991 dd->ipath_pd[port] = NULL;
1992 ipath_free_pddata(dd, pd);
1993 }
1994 kfree(dd->ipath_pd);
1995 /*
1996 * debuggability, in case some cleanup path tries to use it
1997 * after this
1998 */
1999 dd->ipath_pd = NULL;
2000}
2001
2002static void __exit infinipath_cleanup(void) 2033static void __exit infinipath_cleanup(void)
2003{ 2034{
2004 struct ipath_devdata *dd, *tmp;
2005 unsigned long flags;
2006
2007 ipath_diagpkt_remove();
2008
2009 ipath_exit_ipathfs(); 2035 ipath_exit_ipathfs();
2010 2036
2011 ipath_driver_remove_group(&ipath_driver.driver); 2037 ipath_driver_remove_group(&ipath_driver.driver);
2012 2038
2013 spin_lock_irqsave(&ipath_devs_lock, flags);
2014
2015 /*
2016 * turn off rcv, send, and interrupts for all ports, all drivers
2017 * should also hard reset the chip here?
2018 * free up port 0 (kernel) rcvhdr, egr bufs, and eventually tid bufs
2019 * for all versions of the driver, if they were allocated
2020 */
2021 list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) {
2022 spin_unlock_irqrestore(&ipath_devs_lock, flags);
2023
2024 if (dd->ipath_kregbase)
2025 cleanup_device(dd);
2026
2027 if (dd->pcidev) {
2028 if (dd->pcidev->irq) {
2029 ipath_cdbg(VERBOSE,
2030 "unit %u free_irq of irq %x\n",
2031 dd->ipath_unit, dd->pcidev->irq);
2032 free_irq(dd->pcidev->irq, dd);
2033 } else
2034 ipath_dbg("irq is 0, not doing free_irq "
2035 "for unit %u\n", dd->ipath_unit);
2036
2037 /*
2038 * we check for NULL here, because it's outside
2039 * the kregbase check, and we need to call it
2040 * after the free_irq. Thus it's possible that
2041 * the function pointers were never initialized.
2042 */
2043 if (dd->ipath_f_cleanup)
2044 /* clean up chip-specific stuff */
2045 dd->ipath_f_cleanup(dd);
2046
2047 dd->pcidev = NULL;
2048 }
2049 spin_lock_irqsave(&ipath_devs_lock, flags);
2050 }
2051
2052 spin_unlock_irqrestore(&ipath_devs_lock, flags);
2053
2054 ipath_cdbg(VERBOSE, "Unregistering pci driver\n"); 2039 ipath_cdbg(VERBOSE, "Unregistering pci driver\n");
2055 pci_unregister_driver(&ipath_driver); 2040 pci_unregister_driver(&ipath_driver);
2056 2041
diff --git a/drivers/infiniband/hw/ipath/ipath_eeprom.c b/drivers/infiniband/hw/ipath/ipath_eeprom.c
index 3313356ab93a..a4019a6b7560 100644
--- a/drivers/infiniband/hw/ipath/ipath_eeprom.c
+++ b/drivers/infiniband/hw/ipath/ipath_eeprom.c
@@ -100,9 +100,9 @@ static int i2c_gpio_set(struct ipath_devdata *dd,
100 gpioval = &dd->ipath_gpio_out; 100 gpioval = &dd->ipath_gpio_out;
101 read_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extctrl); 101 read_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extctrl);
102 if (line == i2c_line_scl) 102 if (line == i2c_line_scl)
103 mask = ipath_gpio_scl; 103 mask = dd->ipath_gpio_scl;
104 else 104 else
105 mask = ipath_gpio_sda; 105 mask = dd->ipath_gpio_sda;
106 106
107 if (new_line_state == i2c_line_high) 107 if (new_line_state == i2c_line_high)
108 /* tri-state the output rather than force high */ 108 /* tri-state the output rather than force high */
@@ -119,12 +119,12 @@ static int i2c_gpio_set(struct ipath_devdata *dd,
119 write_val = 0x0UL; 119 write_val = 0x0UL;
120 120
121 if (line == i2c_line_scl) { 121 if (line == i2c_line_scl) {
122 write_val <<= ipath_gpio_scl_num; 122 write_val <<= dd->ipath_gpio_scl_num;
123 *gpioval = *gpioval & ~(1UL << ipath_gpio_scl_num); 123 *gpioval = *gpioval & ~(1UL << dd->ipath_gpio_scl_num);
124 *gpioval |= write_val; 124 *gpioval |= write_val;
125 } else { 125 } else {
126 write_val <<= ipath_gpio_sda_num; 126 write_val <<= dd->ipath_gpio_sda_num;
127 *gpioval = *gpioval & ~(1UL << ipath_gpio_sda_num); 127 *gpioval = *gpioval & ~(1UL << dd->ipath_gpio_sda_num);
128 *gpioval |= write_val; 128 *gpioval |= write_val;
129 } 129 }
130 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_out, *gpioval); 130 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_out, *gpioval);
@@ -157,9 +157,9 @@ static int i2c_gpio_get(struct ipath_devdata *dd,
157 read_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extctrl); 157 read_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extctrl);
158 /* config line to be an input */ 158 /* config line to be an input */
159 if (line == i2c_line_scl) 159 if (line == i2c_line_scl)
160 mask = ipath_gpio_scl; 160 mask = dd->ipath_gpio_scl;
161 else 161 else
162 mask = ipath_gpio_sda; 162 mask = dd->ipath_gpio_sda;
163 write_val = read_val & ~mask; 163 write_val = read_val & ~mask;
164 ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, write_val); 164 ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, write_val);
165 read_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extstatus); 165 read_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extstatus);
@@ -187,6 +187,7 @@ bail:
187static void i2c_wait_for_writes(struct ipath_devdata *dd) 187static void i2c_wait_for_writes(struct ipath_devdata *dd)
188{ 188{
189 (void)ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); 189 (void)ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
190 rmb();
190} 191}
191 192
192static void scl_out(struct ipath_devdata *dd, u8 bit) 193static void scl_out(struct ipath_devdata *dd, u8 bit)
diff --git a/drivers/infiniband/hw/ipath/ipath_file_ops.c b/drivers/infiniband/hw/ipath/ipath_file_ops.c
index 29930e22318e..a9ddc6911f66 100644
--- a/drivers/infiniband/hw/ipath/ipath_file_ops.c
+++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c
@@ -41,6 +41,12 @@
41#include "ipath_kernel.h" 41#include "ipath_kernel.h"
42#include "ipath_common.h" 42#include "ipath_common.h"
43 43
44/*
45 * mmap64 doesn't allow all 64 bits for 32-bit applications
46 * so only use the low 43 bits.
47 */
48#define MMAP64_MASK 0x7FFFFFFFFFFUL
49
44static int ipath_open(struct inode *, struct file *); 50static int ipath_open(struct inode *, struct file *);
45static int ipath_close(struct inode *, struct file *); 51static int ipath_close(struct inode *, struct file *);
46static ssize_t ipath_write(struct file *, const char __user *, size_t, 52static ssize_t ipath_write(struct file *, const char __user *, size_t,
@@ -57,18 +63,35 @@ static struct file_operations ipath_file_ops = {
57 .mmap = ipath_mmap 63 .mmap = ipath_mmap
58}; 64};
59 65
60static int ipath_get_base_info(struct ipath_portdata *pd, 66static int ipath_get_base_info(struct file *fp,
61 void __user *ubase, size_t ubase_size) 67 void __user *ubase, size_t ubase_size)
62{ 68{
69 struct ipath_portdata *pd = port_fp(fp);
63 int ret = 0; 70 int ret = 0;
64 struct ipath_base_info *kinfo = NULL; 71 struct ipath_base_info *kinfo = NULL;
65 struct ipath_devdata *dd = pd->port_dd; 72 struct ipath_devdata *dd = pd->port_dd;
73 unsigned subport_cnt;
74 int shared, master;
75 size_t sz;
76
77 subport_cnt = pd->port_subport_cnt;
78 if (!subport_cnt) {
79 shared = 0;
80 master = 0;
81 subport_cnt = 1;
82 } else {
83 shared = 1;
84 master = !subport_fp(fp);
85 }
66 86
67 if (ubase_size < sizeof(*kinfo)) { 87 sz = sizeof(*kinfo);
88 /* If port sharing is not requested, allow the old size structure */
89 if (!shared)
90 sz -= 3 * sizeof(u64);
91 if (ubase_size < sz) {
68 ipath_cdbg(PROC, 92 ipath_cdbg(PROC,
69 "Base size %lu, need %lu (version mismatch?)\n", 93 "Base size %zu, need %zu (version mismatch?)\n",
70 (unsigned long) ubase_size, 94 ubase_size, sz);
71 (unsigned long) sizeof(*kinfo));
72 ret = -EINVAL; 95 ret = -EINVAL;
73 goto bail; 96 goto bail;
74 } 97 }
@@ -95,7 +118,9 @@ static int ipath_get_base_info(struct ipath_portdata *pd,
95 kinfo->spi_rcv_egrperchunk = pd->port_rcvegrbufs_perchunk; 118 kinfo->spi_rcv_egrperchunk = pd->port_rcvegrbufs_perchunk;
96 kinfo->spi_rcv_egrchunksize = kinfo->spi_rcv_egrbuftotlen / 119 kinfo->spi_rcv_egrchunksize = kinfo->spi_rcv_egrbuftotlen /
97 pd->port_rcvegrbuf_chunks; 120 pd->port_rcvegrbuf_chunks;
98 kinfo->spi_tidcnt = dd->ipath_rcvtidcnt; 121 kinfo->spi_tidcnt = dd->ipath_rcvtidcnt / subport_cnt;
122 if (master)
123 kinfo->spi_tidcnt += dd->ipath_rcvtidcnt % subport_cnt;
99 /* 124 /*
100 * for this use, may be ipath_cfgports summed over all chips that 125 * for this use, may be ipath_cfgports summed over all chips that
101 * are are configured and present 126 * are are configured and present
@@ -118,31 +143,75 @@ static int ipath_get_base_info(struct ipath_portdata *pd,
118 * page_address() macro worked, but in 2.6.11, even that returns the 143 * page_address() macro worked, but in 2.6.11, even that returns the
119 * full 64 bit address (upper bits all 1's). So far, using the 144 * full 64 bit address (upper bits all 1's). So far, using the
120 * physical addresses (or chip offsets, for chip mapping) works, but 145 * physical addresses (or chip offsets, for chip mapping) works, but
121 * no doubt some future kernel release will chang that, and we'll be 146 * no doubt some future kernel release will change that, and we'll be
122 * on to yet another method of dealing with this 147 * on to yet another method of dealing with this.
123 */ 148 */
124 kinfo->spi_rcvhdr_base = (u64) pd->port_rcvhdrq_phys; 149 kinfo->spi_rcvhdr_base = (u64) pd->port_rcvhdrq_phys;
125 kinfo->spi_rcvhdr_tailaddr = (u64)pd->port_rcvhdrqtailaddr_phys; 150 kinfo->spi_rcvhdr_tailaddr = (u64) pd->port_rcvhdrqtailaddr_phys;
126 kinfo->spi_rcv_egrbufs = (u64) pd->port_rcvegr_phys; 151 kinfo->spi_rcv_egrbufs = (u64) pd->port_rcvegr_phys;
127 kinfo->spi_pioavailaddr = (u64) dd->ipath_pioavailregs_phys; 152 kinfo->spi_pioavailaddr = (u64) dd->ipath_pioavailregs_phys;
128 kinfo->spi_status = (u64) kinfo->spi_pioavailaddr + 153 kinfo->spi_status = (u64) kinfo->spi_pioavailaddr +
129 (void *) dd->ipath_statusp - 154 (void *) dd->ipath_statusp -
130 (void *) dd->ipath_pioavailregs_dma; 155 (void *) dd->ipath_pioavailregs_dma;
131 kinfo->spi_piobufbase = (u64) pd->port_piobufs; 156 if (!shared) {
132 kinfo->__spi_uregbase = 157 kinfo->spi_piocnt = dd->ipath_pbufsport;
133 dd->ipath_uregbase + dd->ipath_palign * pd->port_port; 158 kinfo->spi_piobufbase = (u64) pd->port_piobufs;
159 kinfo->__spi_uregbase = (u64) dd->ipath_uregbase +
160 dd->ipath_palign * pd->port_port;
161 } else if (master) {
162 kinfo->spi_piocnt = (dd->ipath_pbufsport / subport_cnt) +
163 (dd->ipath_pbufsport % subport_cnt);
164 /* Master's PIO buffers are after all the slave's */
165 kinfo->spi_piobufbase = (u64) pd->port_piobufs +
166 dd->ipath_palign *
167 (dd->ipath_pbufsport - kinfo->spi_piocnt);
168 kinfo->__spi_uregbase = (u64) dd->ipath_uregbase +
169 dd->ipath_palign * pd->port_port;
170 } else {
171 unsigned slave = subport_fp(fp) - 1;
172
173 kinfo->spi_piocnt = dd->ipath_pbufsport / subport_cnt;
174 kinfo->spi_piobufbase = (u64) pd->port_piobufs +
175 dd->ipath_palign * kinfo->spi_piocnt * slave;
176 kinfo->__spi_uregbase = ((u64) pd->subport_uregbase +
177 PAGE_SIZE * slave) & MMAP64_MASK;
134 178
135 kinfo->spi_pioindex = dd->ipath_pbufsport * (pd->port_port - 1); 179 kinfo->spi_rcvhdr_base = ((u64) pd->subport_rcvhdr_base +
136 kinfo->spi_piocnt = dd->ipath_pbufsport; 180 pd->port_rcvhdrq_size * slave) & MMAP64_MASK;
181 kinfo->spi_rcvhdr_tailaddr =
182 (u64) pd->port_rcvhdrqtailaddr_phys & MMAP64_MASK;
183 kinfo->spi_rcv_egrbufs = ((u64) pd->subport_rcvegrbuf +
184 dd->ipath_rcvegrcnt * dd->ipath_rcvegrbufsize * slave) &
185 MMAP64_MASK;
186 }
187
188 kinfo->spi_pioindex = (kinfo->spi_piobufbase - dd->ipath_piobufbase) /
189 dd->ipath_palign;
137 kinfo->spi_pioalign = dd->ipath_palign; 190 kinfo->spi_pioalign = dd->ipath_palign;
138 191
139 kinfo->spi_qpair = IPATH_KD_QP; 192 kinfo->spi_qpair = IPATH_KD_QP;
140 kinfo->spi_piosize = dd->ipath_ibmaxlen; 193 kinfo->spi_piosize = dd->ipath_ibmaxlen;
141 kinfo->spi_mtu = dd->ipath_ibmaxlen; /* maxlen, not ibmtu */ 194 kinfo->spi_mtu = dd->ipath_ibmaxlen; /* maxlen, not ibmtu */
142 kinfo->spi_port = pd->port_port; 195 kinfo->spi_port = pd->port_port;
196 kinfo->spi_subport = subport_fp(fp);
143 kinfo->spi_sw_version = IPATH_KERN_SWVERSION; 197 kinfo->spi_sw_version = IPATH_KERN_SWVERSION;
144 kinfo->spi_hw_version = dd->ipath_revision; 198 kinfo->spi_hw_version = dd->ipath_revision;
145 199
200 if (master) {
201 kinfo->spi_runtime_flags |= IPATH_RUNTIME_MASTER;
202 kinfo->spi_subport_uregbase =
203 (u64) pd->subport_uregbase & MMAP64_MASK;
204 kinfo->spi_subport_rcvegrbuf =
205 (u64) pd->subport_rcvegrbuf & MMAP64_MASK;
206 kinfo->spi_subport_rcvhdr_base =
207 (u64) pd->subport_rcvhdr_base & MMAP64_MASK;
208 ipath_cdbg(PROC, "port %u flags %x %llx %llx %llx\n",
209 kinfo->spi_port, kinfo->spi_runtime_flags,
210 (unsigned long long) kinfo->spi_subport_uregbase,
211 (unsigned long long) kinfo->spi_subport_rcvegrbuf,
212 (unsigned long long) kinfo->spi_subport_rcvhdr_base);
213 }
214
146 if (copy_to_user(ubase, kinfo, sizeof(*kinfo))) 215 if (copy_to_user(ubase, kinfo, sizeof(*kinfo)))
147 ret = -EFAULT; 216 ret = -EFAULT;
148 217
@@ -154,6 +223,7 @@ bail:
154/** 223/**
155 * ipath_tid_update - update a port TID 224 * ipath_tid_update - update a port TID
156 * @pd: the port 225 * @pd: the port
226 * @fp: the ipath device file
157 * @ti: the TID information 227 * @ti: the TID information
158 * 228 *
159 * The new implementation as of Oct 2004 is that the driver assigns 229 * The new implementation as of Oct 2004 is that the driver assigns
@@ -176,11 +246,11 @@ bail:
176 * virtually contiguous pages, that should change to improve 246 * virtually contiguous pages, that should change to improve
177 * performance. 247 * performance.
178 */ 248 */
179static int ipath_tid_update(struct ipath_portdata *pd, 249static int ipath_tid_update(struct ipath_portdata *pd, struct file *fp,
180 const struct ipath_tid_info *ti) 250 const struct ipath_tid_info *ti)
181{ 251{
182 int ret = 0, ntids; 252 int ret = 0, ntids;
183 u32 tid, porttid, cnt, i, tidcnt; 253 u32 tid, porttid, cnt, i, tidcnt, tidoff;
184 u16 *tidlist; 254 u16 *tidlist;
185 struct ipath_devdata *dd = pd->port_dd; 255 struct ipath_devdata *dd = pd->port_dd;
186 u64 physaddr; 256 u64 physaddr;
@@ -188,6 +258,7 @@ static int ipath_tid_update(struct ipath_portdata *pd,
188 u64 __iomem *tidbase; 258 u64 __iomem *tidbase;
189 unsigned long tidmap[8]; 259 unsigned long tidmap[8];
190 struct page **pagep = NULL; 260 struct page **pagep = NULL;
261 unsigned subport = subport_fp(fp);
191 262
192 if (!dd->ipath_pageshadow) { 263 if (!dd->ipath_pageshadow) {
193 ret = -ENOMEM; 264 ret = -ENOMEM;
@@ -204,20 +275,34 @@ static int ipath_tid_update(struct ipath_portdata *pd,
204 ret = -EFAULT; 275 ret = -EFAULT;
205 goto done; 276 goto done;
206 } 277 }
207 tidcnt = dd->ipath_rcvtidcnt; 278 porttid = pd->port_port * dd->ipath_rcvtidcnt;
208 if (cnt >= tidcnt) { 279 if (!pd->port_subport_cnt) {
280 tidcnt = dd->ipath_rcvtidcnt;
281 tid = pd->port_tidcursor;
282 tidoff = 0;
283 } else if (!subport) {
284 tidcnt = (dd->ipath_rcvtidcnt / pd->port_subport_cnt) +
285 (dd->ipath_rcvtidcnt % pd->port_subport_cnt);
286 tidoff = dd->ipath_rcvtidcnt - tidcnt;
287 porttid += tidoff;
288 tid = tidcursor_fp(fp);
289 } else {
290 tidcnt = dd->ipath_rcvtidcnt / pd->port_subport_cnt;
291 tidoff = tidcnt * (subport - 1);
292 porttid += tidoff;
293 tid = tidcursor_fp(fp);
294 }
295 if (cnt > tidcnt) {
209 /* make sure it all fits in port_tid_pg_list */ 296 /* make sure it all fits in port_tid_pg_list */
210 dev_info(&dd->pcidev->dev, "Process tried to allocate %u " 297 dev_info(&dd->pcidev->dev, "Process tried to allocate %u "
211 "TIDs, only trying max (%u)\n", cnt, tidcnt); 298 "TIDs, only trying max (%u)\n", cnt, tidcnt);
212 cnt = tidcnt; 299 cnt = tidcnt;
213 } 300 }
214 pagep = (struct page **)pd->port_tid_pg_list; 301 pagep = &((struct page **) pd->port_tid_pg_list)[tidoff];
215 tidlist = (u16 *) (&pagep[cnt]); 302 tidlist = &((u16 *) &pagep[dd->ipath_rcvtidcnt])[tidoff];
216 303
217 memset(tidmap, 0, sizeof(tidmap)); 304 memset(tidmap, 0, sizeof(tidmap));
218 tid = pd->port_tidcursor;
219 /* before decrement; chip actual # */ 305 /* before decrement; chip actual # */
220 porttid = pd->port_port * tidcnt;
221 ntids = tidcnt; 306 ntids = tidcnt;
222 tidbase = (u64 __iomem *) (((char __iomem *) dd->ipath_kregbase) + 307 tidbase = (u64 __iomem *) (((char __iomem *) dd->ipath_kregbase) +
223 dd->ipath_rcvtidbase + 308 dd->ipath_rcvtidbase +
@@ -274,16 +359,19 @@ static int ipath_tid_update(struct ipath_portdata *pd,
274 ret = -ENOMEM; 359 ret = -ENOMEM;
275 break; 360 break;
276 } 361 }
277 tidlist[i] = tid; 362 tidlist[i] = tid + tidoff;
278 ipath_cdbg(VERBOSE, "Updating idx %u to TID %u, " 363 ipath_cdbg(VERBOSE, "Updating idx %u to TID %u, "
279 "vaddr %lx\n", i, tid, vaddr); 364 "vaddr %lx\n", i, tid + tidoff, vaddr);
280 /* we "know" system pages and TID pages are same size */ 365 /* we "know" system pages and TID pages are same size */
281 dd->ipath_pageshadow[porttid + tid] = pagep[i]; 366 dd->ipath_pageshadow[porttid + tid] = pagep[i];
367 dd->ipath_physshadow[porttid + tid] = ipath_map_page(
368 dd->pcidev, pagep[i], 0, PAGE_SIZE,
369 PCI_DMA_FROMDEVICE);
282 /* 370 /*
283 * don't need atomic or it's overhead 371 * don't need atomic or it's overhead
284 */ 372 */
285 __set_bit(tid, tidmap); 373 __set_bit(tid, tidmap);
286 physaddr = page_to_phys(pagep[i]); 374 physaddr = dd->ipath_physshadow[porttid + tid];
287 ipath_stats.sps_pagelocks++; 375 ipath_stats.sps_pagelocks++;
288 ipath_cdbg(VERBOSE, 376 ipath_cdbg(VERBOSE,
289 "TID %u, vaddr %lx, physaddr %llx pgp %p\n", 377 "TID %u, vaddr %lx, physaddr %llx pgp %p\n",
@@ -317,6 +405,9 @@ static int ipath_tid_update(struct ipath_portdata *pd,
317 tid); 405 tid);
318 dd->ipath_f_put_tid(dd, &tidbase[tid], 1, 406 dd->ipath_f_put_tid(dd, &tidbase[tid], 1,
319 dd->ipath_tidinvalid); 407 dd->ipath_tidinvalid);
408 pci_unmap_page(dd->pcidev,
409 dd->ipath_physshadow[porttid + tid],
410 PAGE_SIZE, PCI_DMA_FROMDEVICE);
320 dd->ipath_pageshadow[porttid + tid] = NULL; 411 dd->ipath_pageshadow[porttid + tid] = NULL;
321 ipath_stats.sps_pageunlocks++; 412 ipath_stats.sps_pageunlocks++;
322 } 413 }
@@ -341,7 +432,10 @@ static int ipath_tid_update(struct ipath_portdata *pd,
341 } 432 }
342 if (tid == tidcnt) 433 if (tid == tidcnt)
343 tid = 0; 434 tid = 0;
344 pd->port_tidcursor = tid; 435 if (!pd->port_subport_cnt)
436 pd->port_tidcursor = tid;
437 else
438 tidcursor_fp(fp) = tid;
345 } 439 }
346 440
347done: 441done:
@@ -354,6 +448,7 @@ done:
354/** 448/**
355 * ipath_tid_free - free a port TID 449 * ipath_tid_free - free a port TID
356 * @pd: the port 450 * @pd: the port
451 * @subport: the subport
357 * @ti: the TID info 452 * @ti: the TID info
358 * 453 *
359 * right now we are unlocking one page at a time, but since 454 * right now we are unlocking one page at a time, but since
@@ -367,7 +462,7 @@ done:
367 * they pass in to us. 462 * they pass in to us.
368 */ 463 */
369 464
370static int ipath_tid_free(struct ipath_portdata *pd, 465static int ipath_tid_free(struct ipath_portdata *pd, unsigned subport,
371 const struct ipath_tid_info *ti) 466 const struct ipath_tid_info *ti)
372{ 467{
373 int ret = 0; 468 int ret = 0;
@@ -388,11 +483,20 @@ static int ipath_tid_free(struct ipath_portdata *pd,
388 } 483 }
389 484
390 porttid = pd->port_port * dd->ipath_rcvtidcnt; 485 porttid = pd->port_port * dd->ipath_rcvtidcnt;
486 if (!pd->port_subport_cnt)
487 tidcnt = dd->ipath_rcvtidcnt;
488 else if (!subport) {
489 tidcnt = (dd->ipath_rcvtidcnt / pd->port_subport_cnt) +
490 (dd->ipath_rcvtidcnt % pd->port_subport_cnt);
491 porttid += dd->ipath_rcvtidcnt - tidcnt;
492 } else {
493 tidcnt = dd->ipath_rcvtidcnt / pd->port_subport_cnt;
494 porttid += tidcnt * (subport - 1);
495 }
391 tidbase = (u64 __iomem *) ((char __iomem *)(dd->ipath_kregbase) + 496 tidbase = (u64 __iomem *) ((char __iomem *)(dd->ipath_kregbase) +
392 dd->ipath_rcvtidbase + 497 dd->ipath_rcvtidbase +
393 porttid * sizeof(*tidbase)); 498 porttid * sizeof(*tidbase));
394 499
395 tidcnt = dd->ipath_rcvtidcnt;
396 limit = sizeof(tidmap) * BITS_PER_BYTE; 500 limit = sizeof(tidmap) * BITS_PER_BYTE;
397 if (limit > tidcnt) 501 if (limit > tidcnt)
398 /* just in case size changes in future */ 502 /* just in case size changes in future */
@@ -417,6 +521,9 @@ static int ipath_tid_free(struct ipath_portdata *pd,
417 pd->port_pid, tid); 521 pd->port_pid, tid);
418 dd->ipath_f_put_tid(dd, &tidbase[tid], 1, 522 dd->ipath_f_put_tid(dd, &tidbase[tid], 1,
419 dd->ipath_tidinvalid); 523 dd->ipath_tidinvalid);
524 pci_unmap_page(dd->pcidev,
525 dd->ipath_physshadow[porttid + tid],
526 PAGE_SIZE, PCI_DMA_FROMDEVICE);
420 ipath_release_user_pages( 527 ipath_release_user_pages(
421 &dd->ipath_pageshadow[porttid + tid], 1); 528 &dd->ipath_pageshadow[porttid + tid], 1);
422 dd->ipath_pageshadow[porttid + tid] = NULL; 529 dd->ipath_pageshadow[porttid + tid] = NULL;
@@ -581,20 +688,24 @@ bail:
581/** 688/**
582 * ipath_manage_rcvq - manage a port's receive queue 689 * ipath_manage_rcvq - manage a port's receive queue
583 * @pd: the port 690 * @pd: the port
691 * @subport: the subport
584 * @start_stop: action to carry out 692 * @start_stop: action to carry out
585 * 693 *
586 * start_stop == 0 disables receive on the port, for use in queue 694 * start_stop == 0 disables receive on the port, for use in queue
587 * overflow conditions. start_stop==1 re-enables, to be used to 695 * overflow conditions. start_stop==1 re-enables, to be used to
588 * re-init the software copy of the head register 696 * re-init the software copy of the head register
589 */ 697 */
590static int ipath_manage_rcvq(struct ipath_portdata *pd, int start_stop) 698static int ipath_manage_rcvq(struct ipath_portdata *pd, unsigned subport,
699 int start_stop)
591{ 700{
592 struct ipath_devdata *dd = pd->port_dd; 701 struct ipath_devdata *dd = pd->port_dd;
593 u64 tval; 702 u64 tval;
594 703
595 ipath_cdbg(PROC, "%sabling rcv for unit %u port %u\n", 704 ipath_cdbg(PROC, "%sabling rcv for unit %u port %u:%u\n",
596 start_stop ? "en" : "dis", dd->ipath_unit, 705 start_stop ? "en" : "dis", dd->ipath_unit,
597 pd->port_port); 706 pd->port_port, subport);
707 if (subport)
708 goto bail;
598 /* atomically clear receive enable port. */ 709 /* atomically clear receive enable port. */
599 if (start_stop) { 710 if (start_stop) {
600 /* 711 /*
@@ -609,7 +720,7 @@ static int ipath_manage_rcvq(struct ipath_portdata *pd, int start_stop)
609 * updated and correct itself, even in the face of software 720 * updated and correct itself, even in the face of software
610 * bugs. 721 * bugs.
611 */ 722 */
612 *pd->port_rcvhdrtail_kvaddr = 0; 723 *(volatile u64 *)pd->port_rcvhdrtail_kvaddr = 0;
613 set_bit(INFINIPATH_R_PORTENABLE_SHIFT + pd->port_port, 724 set_bit(INFINIPATH_R_PORTENABLE_SHIFT + pd->port_port,
614 &dd->ipath_rcvctrl); 725 &dd->ipath_rcvctrl);
615 } else 726 } else
@@ -630,6 +741,7 @@ static int ipath_manage_rcvq(struct ipath_portdata *pd, int start_stop)
630 tval = ipath_read_ureg32(dd, ur_rcvhdrtail, pd->port_port); 741 tval = ipath_read_ureg32(dd, ur_rcvhdrtail, pd->port_port);
631 } 742 }
632 /* always; new head should be equal to new tail; see above */ 743 /* always; new head should be equal to new tail; see above */
744bail:
633 return 0; 745 return 0;
634} 746}
635 747
@@ -687,6 +799,36 @@ static void ipath_clean_part_key(struct ipath_portdata *pd,
687 } 799 }
688} 800}
689 801
802/*
803 * Initialize the port data with the receive buffer sizes
804 * so this can be done while the master port is locked.
805 * Otherwise, there is a race with a slave opening the port
806 * and seeing these fields uninitialized.
807 */
808static void init_user_egr_sizes(struct ipath_portdata *pd)
809{
810 struct ipath_devdata *dd = pd->port_dd;
811 unsigned egrperchunk, egrcnt, size;
812
813 /*
814 * to avoid wasting a lot of memory, we allocate 32KB chunks of
815 * physically contiguous memory, advance through it until used up
816 * and then allocate more. Of course, we need memory to store those
817 * extra pointers, now. Started out with 256KB, but under heavy
818 * memory pressure (creating large files and then copying them over
819 * NFS while doing lots of MPI jobs), we hit some allocation
820 * failures, even though we can sleep... (2.6.10) Still get
821 * failures at 64K. 32K is the lowest we can go without wasting
822 * additional memory.
823 */
824 size = 0x8000;
825 egrperchunk = size / dd->ipath_rcvegrbufsize;
826 egrcnt = dd->ipath_rcvegrcnt;
827 pd->port_rcvegrbuf_chunks = (egrcnt + egrperchunk - 1) / egrperchunk;
828 pd->port_rcvegrbufs_perchunk = egrperchunk;
829 pd->port_rcvegrbuf_size = size;
830}
831
690/** 832/**
691 * ipath_create_user_egr - allocate eager TID buffers 833 * ipath_create_user_egr - allocate eager TID buffers
692 * @pd: the port to allocate TID buffers for 834 * @pd: the port to allocate TID buffers for
@@ -702,7 +844,7 @@ static void ipath_clean_part_key(struct ipath_portdata *pd,
702static int ipath_create_user_egr(struct ipath_portdata *pd) 844static int ipath_create_user_egr(struct ipath_portdata *pd)
703{ 845{
704 struct ipath_devdata *dd = pd->port_dd; 846 struct ipath_devdata *dd = pd->port_dd;
705 unsigned e, egrcnt, alloced, egrperchunk, chunk, egrsize, egroff; 847 unsigned e, egrcnt, egrperchunk, chunk, egrsize, egroff;
706 size_t size; 848 size_t size;
707 int ret; 849 int ret;
708 gfp_t gfp_flags; 850 gfp_t gfp_flags;
@@ -722,31 +864,18 @@ static int ipath_create_user_egr(struct ipath_portdata *pd)
722 ipath_cdbg(VERBOSE, "Allocating %d egr buffers, at egrtid " 864 ipath_cdbg(VERBOSE, "Allocating %d egr buffers, at egrtid "
723 "offset %x, egrsize %u\n", egrcnt, egroff, egrsize); 865 "offset %x, egrsize %u\n", egrcnt, egroff, egrsize);
724 866
725 /* 867 chunk = pd->port_rcvegrbuf_chunks;
726 * to avoid wasting a lot of memory, we allocate 32KB chunks of 868 egrperchunk = pd->port_rcvegrbufs_perchunk;
727 * physically contiguous memory, advance through it until used up 869 size = pd->port_rcvegrbuf_size;
728 * and then allocate more. Of course, we need memory to store those 870 pd->port_rcvegrbuf = kmalloc(chunk * sizeof(pd->port_rcvegrbuf[0]),
729 * extra pointers, now. Started out with 256KB, but under heavy 871 GFP_KERNEL);
730 * memory pressure (creating large files and then copying them over
731 * NFS while doing lots of MPI jobs), we hit some allocation
732 * failures, even though we can sleep... (2.6.10) Still get
733 * failures at 64K. 32K is the lowest we can go without wasting
734 * additional memory.
735 */
736 size = 0x8000;
737 alloced = ALIGN(egrsize * egrcnt, size);
738 egrperchunk = size / egrsize;
739 chunk = (egrcnt + egrperchunk - 1) / egrperchunk;
740 pd->port_rcvegrbuf_chunks = chunk;
741 pd->port_rcvegrbufs_perchunk = egrperchunk;
742 pd->port_rcvegrbuf_size = size;
743 pd->port_rcvegrbuf = vmalloc(chunk * sizeof(pd->port_rcvegrbuf[0]));
744 if (!pd->port_rcvegrbuf) { 872 if (!pd->port_rcvegrbuf) {
745 ret = -ENOMEM; 873 ret = -ENOMEM;
746 goto bail; 874 goto bail;
747 } 875 }
748 pd->port_rcvegrbuf_phys = 876 pd->port_rcvegrbuf_phys =
749 vmalloc(chunk * sizeof(pd->port_rcvegrbuf_phys[0])); 877 kmalloc(chunk * sizeof(pd->port_rcvegrbuf_phys[0]),
878 GFP_KERNEL);
750 if (!pd->port_rcvegrbuf_phys) { 879 if (!pd->port_rcvegrbuf_phys) {
751 ret = -ENOMEM; 880 ret = -ENOMEM;
752 goto bail_rcvegrbuf; 881 goto bail_rcvegrbuf;
@@ -791,105 +920,23 @@ bail_rcvegrbuf_phys:
791 pd->port_rcvegrbuf_phys[e]); 920 pd->port_rcvegrbuf_phys[e]);
792 921
793 } 922 }
794 vfree(pd->port_rcvegrbuf_phys); 923 kfree(pd->port_rcvegrbuf_phys);
795 pd->port_rcvegrbuf_phys = NULL; 924 pd->port_rcvegrbuf_phys = NULL;
796bail_rcvegrbuf: 925bail_rcvegrbuf:
797 vfree(pd->port_rcvegrbuf); 926 kfree(pd->port_rcvegrbuf);
798 pd->port_rcvegrbuf = NULL; 927 pd->port_rcvegrbuf = NULL;
799bail: 928bail:
800 return ret; 929 return ret;
801} 930}
802 931
803static int ipath_do_user_init(struct ipath_portdata *pd,
804 const struct ipath_user_info *uinfo)
805{
806 int ret = 0;
807 struct ipath_devdata *dd = pd->port_dd;
808 u32 head32;
809
810 /* for now, if major version is different, bail */
811 if ((uinfo->spu_userversion >> 16) != IPATH_USER_SWMAJOR) {
812 dev_info(&dd->pcidev->dev,
813 "User major version %d not same as driver "
814 "major %d\n", uinfo->spu_userversion >> 16,
815 IPATH_USER_SWMAJOR);
816 ret = -ENODEV;
817 goto done;
818 }
819
820 if ((uinfo->spu_userversion & 0xffff) != IPATH_USER_SWMINOR)
821 ipath_dbg("User minor version %d not same as driver "
822 "minor %d\n", uinfo->spu_userversion & 0xffff,
823 IPATH_USER_SWMINOR);
824
825 if (uinfo->spu_rcvhdrsize) {
826 ret = ipath_setrcvhdrsize(dd, uinfo->spu_rcvhdrsize);
827 if (ret)
828 goto done;
829 }
830
831 /* for now we do nothing with rcvhdrcnt: uinfo->spu_rcvhdrcnt */
832
833 /* for right now, kernel piobufs are at end, so port 1 is at 0 */
834 pd->port_piobufs = dd->ipath_piobufbase +
835 dd->ipath_pbufsport * (pd->port_port -
836 1) * dd->ipath_palign;
837 ipath_cdbg(VERBOSE, "Set base of piobufs for port %u to 0x%x\n",
838 pd->port_port, pd->port_piobufs);
839
840 /*
841 * Now allocate the rcvhdr Q and eager TIDs; skip the TID
842 * array for time being. If pd->port_port > chip-supported,
843 * we need to do extra stuff here to handle by handling overflow
844 * through port 0, someday
845 */
846 ret = ipath_create_rcvhdrq(dd, pd);
847 if (!ret)
848 ret = ipath_create_user_egr(pd);
849 if (ret)
850 goto done;
851
852 /*
853 * set the eager head register for this port to the current values
854 * of the tail pointers, since we don't know if they were
855 * updated on last use of the port.
856 */
857 head32 = ipath_read_ureg32(dd, ur_rcvegrindextail, pd->port_port);
858 ipath_write_ureg(dd, ur_rcvegrindexhead, head32, pd->port_port);
859 dd->ipath_lastegrheads[pd->port_port] = -1;
860 dd->ipath_lastrcvhdrqtails[pd->port_port] = -1;
861 ipath_cdbg(VERBOSE, "Wrote port%d egrhead %x from tail regs\n",
862 pd->port_port, head32);
863 pd->port_tidcursor = 0; /* start at beginning after open */
864 /*
865 * now enable the port; the tail registers will be written to memory
866 * by the chip as soon as it sees the write to
867 * dd->ipath_kregs->kr_rcvctrl. The update only happens on
868 * transition from 0 to 1, so clear it first, then set it as part of
869 * enabling the port. This will (very briefly) affect any other
870 * open ports, but it shouldn't be long enough to be an issue.
871 * We explictly set the in-memory copy to 0 beforehand, so we don't
872 * have to wait to be sure the DMA update has happened.
873 */
874 *pd->port_rcvhdrtail_kvaddr = 0ULL;
875 set_bit(INFINIPATH_R_PORTENABLE_SHIFT + pd->port_port,
876 &dd->ipath_rcvctrl);
877 ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
878 dd->ipath_rcvctrl & ~INFINIPATH_R_TAILUPD);
879 ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
880 dd->ipath_rcvctrl);
881done:
882 return ret;
883}
884
885 932
886/* common code for the mappings on dma_alloc_coherent mem */ 933/* common code for the mappings on dma_alloc_coherent mem */
887static int ipath_mmap_mem(struct vm_area_struct *vma, 934static int ipath_mmap_mem(struct vm_area_struct *vma,
888 struct ipath_portdata *pd, unsigned len, 935 struct ipath_portdata *pd, unsigned len, int write_ok,
889 int write_ok, dma_addr_t addr, char *what) 936 void *kvaddr, char *what)
890{ 937{
891 struct ipath_devdata *dd = pd->port_dd; 938 struct ipath_devdata *dd = pd->port_dd;
892 unsigned pfn = (unsigned long)addr >> PAGE_SHIFT; 939 unsigned long pfn;
893 int ret; 940 int ret;
894 941
895 if ((vma->vm_end - vma->vm_start) > len) { 942 if ((vma->vm_end - vma->vm_start) > len) {
@@ -912,17 +959,17 @@ static int ipath_mmap_mem(struct vm_area_struct *vma,
912 vma->vm_flags &= ~VM_MAYWRITE; 959 vma->vm_flags &= ~VM_MAYWRITE;
913 } 960 }
914 961
962 pfn = virt_to_phys(kvaddr) >> PAGE_SHIFT;
915 ret = remap_pfn_range(vma, vma->vm_start, pfn, 963 ret = remap_pfn_range(vma, vma->vm_start, pfn,
916 len, vma->vm_page_prot); 964 len, vma->vm_page_prot);
917 if (ret) 965 if (ret)
918 dev_info(&dd->pcidev->dev, 966 dev_info(&dd->pcidev->dev, "%s port%u mmap of %lx, %x "
919 "%s port%u mmap of %lx, %x bytes r%c failed: %d\n", 967 "bytes r%c failed: %d\n", what, pd->port_port,
920 what, pd->port_port, (unsigned long)addr, len, 968 pfn, len, write_ok?'w':'o', ret);
921 write_ok?'w':'o', ret);
922 else 969 else
923 ipath_cdbg(VERBOSE, "%s port%u mmaped %lx, %x bytes r%c\n", 970 ipath_cdbg(VERBOSE, "%s port%u mmaped %lx, %x bytes "
924 what, pd->port_port, (unsigned long)addr, len, 971 "r%c\n", what, pd->port_port, pfn, len,
925 write_ok?'w':'o'); 972 write_ok?'w':'o');
926bail: 973bail:
927 return ret; 974 return ret;
928} 975}
@@ -957,7 +1004,8 @@ static int mmap_ureg(struct vm_area_struct *vma, struct ipath_devdata *dd,
957 1004
958static int mmap_piobufs(struct vm_area_struct *vma, 1005static int mmap_piobufs(struct vm_area_struct *vma,
959 struct ipath_devdata *dd, 1006 struct ipath_devdata *dd,
960 struct ipath_portdata *pd) 1007 struct ipath_portdata *pd,
1008 unsigned piobufs, unsigned piocnt)
961{ 1009{
962 unsigned long phys; 1010 unsigned long phys;
963 int ret; 1011 int ret;
@@ -968,16 +1016,15 @@ static int mmap_piobufs(struct vm_area_struct *vma,
968 * process data, and catches users who might try to read the i/o 1016 * process data, and catches users who might try to read the i/o
969 * space due to a bug. 1017 * space due to a bug.
970 */ 1018 */
971 if ((vma->vm_end - vma->vm_start) > 1019 if ((vma->vm_end - vma->vm_start) > (piocnt * dd->ipath_palign)) {
972 (dd->ipath_pbufsport * dd->ipath_palign)) {
973 dev_info(&dd->pcidev->dev, "FAIL mmap piobufs: " 1020 dev_info(&dd->pcidev->dev, "FAIL mmap piobufs: "
974 "reqlen %lx > PAGE\n", 1021 "reqlen %lx > PAGE\n",
975 vma->vm_end - vma->vm_start); 1022 vma->vm_end - vma->vm_start);
976 ret = -EFAULT; 1023 ret = -EINVAL;
977 goto bail; 1024 goto bail;
978 } 1025 }
979 1026
980 phys = dd->ipath_physaddr + pd->port_piobufs; 1027 phys = dd->ipath_physaddr + piobufs;
981 1028
982 /* 1029 /*
983 * Don't mark this as non-cached, or we don't get the 1030 * Don't mark this as non-cached, or we don't get the
@@ -1011,7 +1058,7 @@ static int mmap_rcvegrbufs(struct vm_area_struct *vma,
1011 struct ipath_devdata *dd = pd->port_dd; 1058 struct ipath_devdata *dd = pd->port_dd;
1012 unsigned long start, size; 1059 unsigned long start, size;
1013 size_t total_size, i; 1060 size_t total_size, i;
1014 dma_addr_t *phys; 1061 unsigned long pfn;
1015 int ret; 1062 int ret;
1016 1063
1017 size = pd->port_rcvegrbuf_size; 1064 size = pd->port_rcvegrbuf_size;
@@ -1021,7 +1068,7 @@ static int mmap_rcvegrbufs(struct vm_area_struct *vma,
1021 "reqlen %lx > actual %lx\n", 1068 "reqlen %lx > actual %lx\n",
1022 vma->vm_end - vma->vm_start, 1069 vma->vm_end - vma->vm_start,
1023 (unsigned long) total_size); 1070 (unsigned long) total_size);
1024 ret = -EFAULT; 1071 ret = -EINVAL;
1025 goto bail; 1072 goto bail;
1026 } 1073 }
1027 1074
@@ -1035,11 +1082,11 @@ static int mmap_rcvegrbufs(struct vm_area_struct *vma,
1035 vma->vm_flags &= ~VM_MAYWRITE; 1082 vma->vm_flags &= ~VM_MAYWRITE;
1036 1083
1037 start = vma->vm_start; 1084 start = vma->vm_start;
1038 phys = pd->port_rcvegrbuf_phys;
1039 1085
1040 for (i = 0; i < pd->port_rcvegrbuf_chunks; i++, start += size) { 1086 for (i = 0; i < pd->port_rcvegrbuf_chunks; i++, start += size) {
1041 ret = remap_pfn_range(vma, start, phys[i] >> PAGE_SHIFT, 1087 pfn = virt_to_phys(pd->port_rcvegrbuf[i]) >> PAGE_SHIFT;
1042 size, vma->vm_page_prot); 1088 ret = remap_pfn_range(vma, start, pfn, size,
1089 vma->vm_page_prot);
1043 if (ret < 0) 1090 if (ret < 0)
1044 goto bail; 1091 goto bail;
1045 } 1092 }
@@ -1049,6 +1096,122 @@ bail:
1049 return ret; 1096 return ret;
1050} 1097}
1051 1098
1099/*
1100 * ipath_file_vma_nopage - handle a VMA page fault.
1101 */
1102static struct page *ipath_file_vma_nopage(struct vm_area_struct *vma,
1103 unsigned long address, int *type)
1104{
1105 unsigned long offset = address - vma->vm_start;
1106 struct page *page = NOPAGE_SIGBUS;
1107 void *pageptr;
1108
1109 /*
1110 * Convert the vmalloc address into a struct page.
1111 */
1112 pageptr = (void *)(offset + (vma->vm_pgoff << PAGE_SHIFT));
1113 page = vmalloc_to_page(pageptr);
1114 if (!page)
1115 goto out;
1116
1117 /* Increment the reference count. */
1118 get_page(page);
1119 if (type)
1120 *type = VM_FAULT_MINOR;
1121out:
1122 return page;
1123}
1124
1125static struct vm_operations_struct ipath_file_vm_ops = {
1126 .nopage = ipath_file_vma_nopage,
1127};
1128
1129static int mmap_kvaddr(struct vm_area_struct *vma, u64 pgaddr,
1130 struct ipath_portdata *pd, unsigned subport)
1131{
1132 unsigned long len;
1133 struct ipath_devdata *dd;
1134 void *addr;
1135 size_t size;
1136 int ret;
1137
1138 /* If the port is not shared, all addresses should be physical */
1139 if (!pd->port_subport_cnt) {
1140 ret = -EINVAL;
1141 goto bail;
1142 }
1143
1144 dd = pd->port_dd;
1145 size = pd->port_rcvegrbuf_chunks * pd->port_rcvegrbuf_size;
1146
1147 /*
1148 * Master has all the slave uregbase, rcvhdrq, and
1149 * rcvegrbufs mmapped.
1150 */
1151 if (subport == 0) {
1152 unsigned num_slaves = pd->port_subport_cnt - 1;
1153
1154 if (pgaddr == ((u64) pd->subport_uregbase & MMAP64_MASK)) {
1155 addr = pd->subport_uregbase;
1156 size = PAGE_SIZE * num_slaves;
1157 } else if (pgaddr == ((u64) pd->subport_rcvhdr_base &
1158 MMAP64_MASK)) {
1159 addr = pd->subport_rcvhdr_base;
1160 size = pd->port_rcvhdrq_size * num_slaves;
1161 } else if (pgaddr == ((u64) pd->subport_rcvegrbuf &
1162 MMAP64_MASK)) {
1163 addr = pd->subport_rcvegrbuf;
1164 size *= num_slaves;
1165 } else {
1166 ret = -EINVAL;
1167 goto bail;
1168 }
1169 } else if (pgaddr == (((u64) pd->subport_uregbase +
1170 PAGE_SIZE * (subport - 1)) & MMAP64_MASK)) {
1171 addr = pd->subport_uregbase + PAGE_SIZE * (subport - 1);
1172 size = PAGE_SIZE;
1173 } else if (pgaddr == (((u64) pd->subport_rcvhdr_base +
1174 pd->port_rcvhdrq_size * (subport - 1)) &
1175 MMAP64_MASK)) {
1176 addr = pd->subport_rcvhdr_base +
1177 pd->port_rcvhdrq_size * (subport - 1);
1178 size = pd->port_rcvhdrq_size;
1179 } else if (pgaddr == (((u64) pd->subport_rcvegrbuf +
1180 size * (subport - 1)) & MMAP64_MASK)) {
1181 addr = pd->subport_rcvegrbuf + size * (subport - 1);
1182 /* rcvegrbufs are read-only on the slave */
1183 if (vma->vm_flags & VM_WRITE) {
1184 dev_info(&dd->pcidev->dev,
1185 "Can't map eager buffers as "
1186 "writable (flags=%lx)\n", vma->vm_flags);
1187 ret = -EPERM;
1188 goto bail;
1189 }
1190 /*
1191 * Don't allow permission to later change to writeable
1192 * with mprotect.
1193 */
1194 vma->vm_flags &= ~VM_MAYWRITE;
1195 } else {
1196 ret = -EINVAL;
1197 goto bail;
1198 }
1199 len = vma->vm_end - vma->vm_start;
1200 if (len > size) {
1201 ipath_cdbg(MM, "FAIL: reqlen %lx > %zx\n", len, size);
1202 ret = -EINVAL;
1203 goto bail;
1204 }
1205
1206 vma->vm_pgoff = (unsigned long) addr >> PAGE_SHIFT;
1207 vma->vm_ops = &ipath_file_vm_ops;
1208 vma->vm_flags |= VM_RESERVED | VM_DONTEXPAND;
1209 ret = 0;
1210
1211bail:
1212 return ret;
1213}
1214
1052/** 1215/**
1053 * ipath_mmap - mmap various structures into user space 1216 * ipath_mmap - mmap various structures into user space
1054 * @fp: the file pointer 1217 * @fp: the file pointer
@@ -1064,73 +1227,99 @@ static int ipath_mmap(struct file *fp, struct vm_area_struct *vma)
1064 struct ipath_portdata *pd; 1227 struct ipath_portdata *pd;
1065 struct ipath_devdata *dd; 1228 struct ipath_devdata *dd;
1066 u64 pgaddr, ureg; 1229 u64 pgaddr, ureg;
1230 unsigned piobufs, piocnt;
1067 int ret; 1231 int ret;
1068 1232
1069 pd = port_fp(fp); 1233 pd = port_fp(fp);
1234 if (!pd) {
1235 ret = -EINVAL;
1236 goto bail;
1237 }
1070 dd = pd->port_dd; 1238 dd = pd->port_dd;
1071 1239
1072 /* 1240 /*
1073 * This is the ipath_do_user_init() code, mapping the shared buffers 1241 * This is the ipath_do_user_init() code, mapping the shared buffers
1074 * into the user process. The address referred to by vm_pgoff is the 1242 * into the user process. The address referred to by vm_pgoff is the
1075 * virtual, not physical, address; we only do one mmap for each 1243 * file offset passed via mmap(). For shared ports, this is the
1076 * space mapped. 1244 * kernel vmalloc() address of the pages to share with the master.
1245 * For non-shared or master ports, this is a physical address.
1246 * We only do one mmap for each space mapped.
1077 */ 1247 */
1078 pgaddr = vma->vm_pgoff << PAGE_SHIFT; 1248 pgaddr = vma->vm_pgoff << PAGE_SHIFT;
1079 1249
1080 /* 1250 /*
1081 * Must fit in 40 bits for our hardware; some checked elsewhere, 1251 * Check for 0 in case one of the allocations failed, but user
1082 * but we'll be paranoid. Check for 0 is mostly in case one of the 1252 * called mmap anyway.
1083 * allocations failed, but user called mmap anyway. We want to catch
1084 * that before it can match.
1085 */ 1253 */
1086 if (!pgaddr || pgaddr >= (1ULL<<40)) { 1254 if (!pgaddr) {
1087 ipath_dev_err(dd, "Bad phys addr %llx, start %lx, end %lx\n", 1255 ret = -EINVAL;
1088 (unsigned long long)pgaddr, vma->vm_start, vma->vm_end); 1256 goto bail;
1089 return -EINVAL;
1090 } 1257 }
1091 1258
1092 /* just the offset of the port user registers, not physical addr */ 1259 ipath_cdbg(MM, "pgaddr %llx vm_start=%lx len %lx port %u:%u:%u\n",
1093 ureg = dd->ipath_uregbase + dd->ipath_palign * pd->port_port;
1094
1095 ipath_cdbg(MM, "ushare: pgaddr %llx vm_start=%lx, vmlen %lx\n",
1096 (unsigned long long) pgaddr, vma->vm_start, 1260 (unsigned long long) pgaddr, vma->vm_start,
1097 vma->vm_end - vma->vm_start); 1261 vma->vm_end - vma->vm_start, dd->ipath_unit,
1262 pd->port_port, subport_fp(fp));
1098 1263
1099 if (vma->vm_start & (PAGE_SIZE-1)) { 1264 /*
1100 ipath_dev_err(dd, 1265 * Physical addresses must fit in 40 bits for our hardware.
1101 "vm_start not aligned: %lx, end=%lx phys %lx\n", 1266 * Check for kernel virtual addresses first, anything else must
1102 vma->vm_start, vma->vm_end, (unsigned long)pgaddr); 1267 * match a HW or memory address.
1103 ret = -EINVAL; 1268 */
1269 if (pgaddr >= (1ULL<<40)) {
1270 ret = mmap_kvaddr(vma, pgaddr, pd, subport_fp(fp));
1271 goto bail;
1104 } 1272 }
1105 else if (pgaddr == ureg) 1273
1274 if (!pd->port_subport_cnt) {
1275 /* port is not shared */
1276 ureg = dd->ipath_uregbase + dd->ipath_palign * pd->port_port;
1277 piocnt = dd->ipath_pbufsport;
1278 piobufs = pd->port_piobufs;
1279 } else if (!subport_fp(fp)) {
1280 /* caller is the master */
1281 ureg = dd->ipath_uregbase + dd->ipath_palign * pd->port_port;
1282 piocnt = (dd->ipath_pbufsport / pd->port_subport_cnt) +
1283 (dd->ipath_pbufsport % pd->port_subport_cnt);
1284 piobufs = pd->port_piobufs +
1285 dd->ipath_palign * (dd->ipath_pbufsport - piocnt);
1286 } else {
1287 unsigned slave = subport_fp(fp) - 1;
1288
1289 /* caller is a slave */
1290 ureg = 0;
1291 piocnt = dd->ipath_pbufsport / pd->port_subport_cnt;
1292 piobufs = pd->port_piobufs + dd->ipath_palign * piocnt * slave;
1293 }
1294
1295 if (pgaddr == ureg)
1106 ret = mmap_ureg(vma, dd, ureg); 1296 ret = mmap_ureg(vma, dd, ureg);
1107 else if (pgaddr == pd->port_piobufs) 1297 else if (pgaddr == piobufs)
1108 ret = mmap_piobufs(vma, dd, pd); 1298 ret = mmap_piobufs(vma, dd, pd, piobufs, piocnt);
1109 else if (pgaddr == (u64) pd->port_rcvegr_phys) 1299 else if (pgaddr == dd->ipath_pioavailregs_phys)
1300 /* in-memory copy of pioavail registers */
1301 ret = ipath_mmap_mem(vma, pd, PAGE_SIZE, 0,
1302 (void *) dd->ipath_pioavailregs_dma,
1303 "pioavail registers");
1304 else if (subport_fp(fp))
1305 /* Subports don't mmap the physical receive buffers */
1306 ret = -EINVAL;
1307 else if (pgaddr == pd->port_rcvegr_phys)
1110 ret = mmap_rcvegrbufs(vma, pd); 1308 ret = mmap_rcvegrbufs(vma, pd);
1111 else if (pgaddr == (u64) pd->port_rcvhdrq_phys) { 1309 else if (pgaddr == (u64) pd->port_rcvhdrq_phys)
1112 /* 1310 /*
1113 * The rcvhdrq itself; readonly except on HT (so have 1311 * The rcvhdrq itself; readonly except on HT (so have
1114 * to allow writable mapping), multiple pages, contiguous 1312 * to allow writable mapping), multiple pages, contiguous
1115 * from an i/o perspective. 1313 * from an i/o perspective.
1116 */ 1314 */
1117 unsigned total_size = 1315 ret = ipath_mmap_mem(vma, pd, pd->port_rcvhdrq_size, 1,
1118 ALIGN(dd->ipath_rcvhdrcnt * dd->ipath_rcvhdrentsize 1316 pd->port_rcvhdrq,
1119 * sizeof(u32), PAGE_SIZE);
1120 ret = ipath_mmap_mem(vma, pd, total_size, 1,
1121 pd->port_rcvhdrq_phys,
1122 "rcvhdrq"); 1317 "rcvhdrq");
1123 } 1318 else if (pgaddr == (u64) pd->port_rcvhdrqtailaddr_phys)
1124 else if (pgaddr == (u64)pd->port_rcvhdrqtailaddr_phys)
1125 /* in-memory copy of rcvhdrq tail register */ 1319 /* in-memory copy of rcvhdrq tail register */
1126 ret = ipath_mmap_mem(vma, pd, PAGE_SIZE, 0, 1320 ret = ipath_mmap_mem(vma, pd, PAGE_SIZE, 0,
1127 pd->port_rcvhdrqtailaddr_phys, 1321 pd->port_rcvhdrtail_kvaddr,
1128 "rcvhdrq tail"); 1322 "rcvhdrq tail");
1129 else if (pgaddr == dd->ipath_pioavailregs_phys)
1130 /* in-memory copy of pioavail registers */
1131 ret = ipath_mmap_mem(vma, pd, PAGE_SIZE, 0,
1132 dd->ipath_pioavailregs_phys,
1133 "pioavail registers");
1134 else 1323 else
1135 ret = -EINVAL; 1324 ret = -EINVAL;
1136 1325
@@ -1138,9 +1327,10 @@ static int ipath_mmap(struct file *fp, struct vm_area_struct *vma)
1138 1327
1139 if (ret < 0) 1328 if (ret < 0)
1140 dev_info(&dd->pcidev->dev, 1329 dev_info(&dd->pcidev->dev,
1141 "Failure %d on addr %lx, off %lx\n", 1330 "Failure %d on off %llx len %lx\n",
1142 -ret, vma->vm_start, vma->vm_pgoff); 1331 -ret, (unsigned long long)pgaddr,
1143 1332 vma->vm_end - vma->vm_start);
1333bail:
1144 return ret; 1334 return ret;
1145} 1335}
1146 1336
@@ -1154,6 +1344,8 @@ static unsigned int ipath_poll(struct file *fp,
1154 struct ipath_devdata *dd; 1344 struct ipath_devdata *dd;
1155 1345
1156 pd = port_fp(fp); 1346 pd = port_fp(fp);
1347 if (!pd)
1348 goto bail;
1157 dd = pd->port_dd; 1349 dd = pd->port_dd;
1158 1350
1159 bit = pd->port_port + INFINIPATH_R_INTRAVAIL_SHIFT; 1351 bit = pd->port_port + INFINIPATH_R_INTRAVAIL_SHIFT;
@@ -1176,7 +1368,7 @@ static unsigned int ipath_poll(struct file *fp,
1176 1368
1177 if (tail == head) { 1369 if (tail == head) {
1178 set_bit(IPATH_PORT_WAITING_RCV, &pd->port_flag); 1370 set_bit(IPATH_PORT_WAITING_RCV, &pd->port_flag);
1179 if(dd->ipath_rhdrhead_intr_off) /* arm rcv interrupt */ 1371 if (dd->ipath_rhdrhead_intr_off) /* arm rcv interrupt */
1180 (void)ipath_write_ureg(dd, ur_rcvhdrhead, 1372 (void)ipath_write_ureg(dd, ur_rcvhdrhead,
1181 dd->ipath_rhdrhead_intr_off 1373 dd->ipath_rhdrhead_intr_off
1182 | head, pd->port_port); 1374 | head, pd->port_port);
@@ -1200,18 +1392,80 @@ static unsigned int ipath_poll(struct file *fp,
1200 ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl, 1392 ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
1201 dd->ipath_rcvctrl); 1393 dd->ipath_rcvctrl);
1202 1394
1395bail:
1203 return pollflag; 1396 return pollflag;
1204} 1397}
1205 1398
1399static int init_subports(struct ipath_devdata *dd,
1400 struct ipath_portdata *pd,
1401 const struct ipath_user_info *uinfo)
1402{
1403 int ret = 0;
1404 unsigned num_slaves;
1405 size_t size;
1406
1407 /* Old user binaries don't know about subports */
1408 if ((uinfo->spu_userversion & 0xffff) != IPATH_USER_SWMINOR)
1409 goto bail;
1410 /*
1411 * If the user is requesting zero or one port,
1412 * skip the subport allocation.
1413 */
1414 if (uinfo->spu_subport_cnt <= 1)
1415 goto bail;
1416 if (uinfo->spu_subport_cnt > 4) {
1417 ret = -EINVAL;
1418 goto bail;
1419 }
1420
1421 num_slaves = uinfo->spu_subport_cnt - 1;
1422 pd->subport_uregbase = vmalloc(PAGE_SIZE * num_slaves);
1423 if (!pd->subport_uregbase) {
1424 ret = -ENOMEM;
1425 goto bail;
1426 }
1427 /* Note: pd->port_rcvhdrq_size isn't initialized yet. */
1428 size = ALIGN(dd->ipath_rcvhdrcnt * dd->ipath_rcvhdrentsize *
1429 sizeof(u32), PAGE_SIZE) * num_slaves;
1430 pd->subport_rcvhdr_base = vmalloc(size);
1431 if (!pd->subport_rcvhdr_base) {
1432 ret = -ENOMEM;
1433 goto bail_ureg;
1434 }
1435
1436 pd->subport_rcvegrbuf = vmalloc(pd->port_rcvegrbuf_chunks *
1437 pd->port_rcvegrbuf_size *
1438 num_slaves);
1439 if (!pd->subport_rcvegrbuf) {
1440 ret = -ENOMEM;
1441 goto bail_rhdr;
1442 }
1443
1444 pd->port_subport_cnt = uinfo->spu_subport_cnt;
1445 pd->port_subport_id = uinfo->spu_subport_id;
1446 pd->active_slaves = 1;
1447 goto bail;
1448
1449bail_rhdr:
1450 vfree(pd->subport_rcvhdr_base);
1451bail_ureg:
1452 vfree(pd->subport_uregbase);
1453 pd->subport_uregbase = NULL;
1454bail:
1455 return ret;
1456}
1457
1206static int try_alloc_port(struct ipath_devdata *dd, int port, 1458static int try_alloc_port(struct ipath_devdata *dd, int port,
1207 struct file *fp) 1459 struct file *fp,
1460 const struct ipath_user_info *uinfo)
1208{ 1461{
1462 struct ipath_portdata *pd;
1209 int ret; 1463 int ret;
1210 1464
1211 if (!dd->ipath_pd[port]) { 1465 if (!(pd = dd->ipath_pd[port])) {
1212 void *p, *ptmp; 1466 void *ptmp;
1213 1467
1214 p = kzalloc(sizeof(struct ipath_portdata), GFP_KERNEL); 1468 pd = kzalloc(sizeof(struct ipath_portdata), GFP_KERNEL);
1215 1469
1216 /* 1470 /*
1217 * Allocate memory for use in ipath_tid_update() just once 1471 * Allocate memory for use in ipath_tid_update() just once
@@ -1221,34 +1475,36 @@ static int try_alloc_port(struct ipath_devdata *dd, int port,
1221 ptmp = kmalloc(dd->ipath_rcvtidcnt * sizeof(u16) + 1475 ptmp = kmalloc(dd->ipath_rcvtidcnt * sizeof(u16) +
1222 dd->ipath_rcvtidcnt * sizeof(struct page **), 1476 dd->ipath_rcvtidcnt * sizeof(struct page **),
1223 GFP_KERNEL); 1477 GFP_KERNEL);
1224 if (!p || !ptmp) { 1478 if (!pd || !ptmp) {
1225 ipath_dev_err(dd, "Unable to allocate portdata " 1479 ipath_dev_err(dd, "Unable to allocate portdata "
1226 "memory, failing open\n"); 1480 "memory, failing open\n");
1227 ret = -ENOMEM; 1481 ret = -ENOMEM;
1228 kfree(p); 1482 kfree(pd);
1229 kfree(ptmp); 1483 kfree(ptmp);
1230 goto bail; 1484 goto bail;
1231 } 1485 }
1232 dd->ipath_pd[port] = p; 1486 dd->ipath_pd[port] = pd;
1233 dd->ipath_pd[port]->port_port = port; 1487 dd->ipath_pd[port]->port_port = port;
1234 dd->ipath_pd[port]->port_dd = dd; 1488 dd->ipath_pd[port]->port_dd = dd;
1235 dd->ipath_pd[port]->port_tid_pg_list = ptmp; 1489 dd->ipath_pd[port]->port_tid_pg_list = ptmp;
1236 init_waitqueue_head(&dd->ipath_pd[port]->port_wait); 1490 init_waitqueue_head(&dd->ipath_pd[port]->port_wait);
1237 } 1491 }
1238 if (!dd->ipath_pd[port]->port_cnt) { 1492 if (!pd->port_cnt) {
1239 dd->ipath_pd[port]->port_cnt = 1; 1493 pd->userversion = uinfo->spu_userversion;
1240 fp->private_data = (void *) dd->ipath_pd[port]; 1494 init_user_egr_sizes(pd);
1495 if ((ret = init_subports(dd, pd, uinfo)) != 0)
1496 goto bail;
1241 ipath_cdbg(PROC, "%s[%u] opened unit:port %u:%u\n", 1497 ipath_cdbg(PROC, "%s[%u] opened unit:port %u:%u\n",
1242 current->comm, current->pid, dd->ipath_unit, 1498 current->comm, current->pid, dd->ipath_unit,
1243 port); 1499 port);
1244 dd->ipath_pd[port]->port_pid = current->pid; 1500 pd->port_cnt = 1;
1245 strncpy(dd->ipath_pd[port]->port_comm, current->comm, 1501 port_fp(fp) = pd;
1246 sizeof(dd->ipath_pd[port]->port_comm)); 1502 pd->port_pid = current->pid;
1503 strncpy(pd->port_comm, current->comm, sizeof(pd->port_comm));
1247 ipath_stats.sps_ports++; 1504 ipath_stats.sps_ports++;
1248 ret = 0; 1505 ret = 0;
1249 goto bail; 1506 } else
1250 } 1507 ret = -EBUSY;
1251 ret = -EBUSY;
1252 1508
1253bail: 1509bail:
1254 return ret; 1510 return ret;
@@ -1264,7 +1520,8 @@ static inline int usable(struct ipath_devdata *dd)
1264 | IPATH_LINKUNK)); 1520 | IPATH_LINKUNK));
1265} 1521}
1266 1522
1267static int find_free_port(int unit, struct file *fp) 1523static int find_free_port(int unit, struct file *fp,
1524 const struct ipath_user_info *uinfo)
1268{ 1525{
1269 struct ipath_devdata *dd = ipath_lookup(unit); 1526 struct ipath_devdata *dd = ipath_lookup(unit);
1270 int ret, i; 1527 int ret, i;
@@ -1279,8 +1536,8 @@ static int find_free_port(int unit, struct file *fp)
1279 goto bail; 1536 goto bail;
1280 } 1537 }
1281 1538
1282 for (i = 0; i < dd->ipath_cfgports; i++) { 1539 for (i = 1; i < dd->ipath_cfgports; i++) {
1283 ret = try_alloc_port(dd, i, fp); 1540 ret = try_alloc_port(dd, i, fp, uinfo);
1284 if (ret != -EBUSY) 1541 if (ret != -EBUSY)
1285 goto bail; 1542 goto bail;
1286 } 1543 }
@@ -1290,13 +1547,14 @@ bail:
1290 return ret; 1547 return ret;
1291} 1548}
1292 1549
1293static int find_best_unit(struct file *fp) 1550static int find_best_unit(struct file *fp,
1551 const struct ipath_user_info *uinfo)
1294{ 1552{
1295 int ret = 0, i, prefunit = -1, devmax; 1553 int ret = 0, i, prefunit = -1, devmax;
1296 int maxofallports, npresent, nup; 1554 int maxofallports, npresent, nup;
1297 int ndev; 1555 int ndev;
1298 1556
1299 (void) ipath_count_units(&npresent, &nup, &maxofallports); 1557 devmax = ipath_count_units(&npresent, &nup, &maxofallports);
1300 1558
1301 /* 1559 /*
1302 * This code is present to allow a knowledgeable person to 1560 * This code is present to allow a knowledgeable person to
@@ -1343,8 +1601,6 @@ static int find_best_unit(struct file *fp)
1343 1601
1344 if (prefunit != -1) 1602 if (prefunit != -1)
1345 devmax = prefunit + 1; 1603 devmax = prefunit + 1;
1346 else
1347 devmax = ipath_count_units(NULL, NULL, NULL);
1348recheck: 1604recheck:
1349 for (i = 1; i < maxofallports; i++) { 1605 for (i = 1; i < maxofallports; i++) {
1350 for (ndev = prefunit != -1 ? prefunit : 0; ndev < devmax; 1606 for (ndev = prefunit != -1 ? prefunit : 0; ndev < devmax;
@@ -1359,7 +1615,7 @@ recheck:
1359 * next. 1615 * next.
1360 */ 1616 */
1361 continue; 1617 continue;
1362 ret = try_alloc_port(dd, i, fp); 1618 ret = try_alloc_port(dd, i, fp, uinfo);
1363 if (!ret) 1619 if (!ret)
1364 goto done; 1620 goto done;
1365 } 1621 }
@@ -1395,22 +1651,183 @@ done:
1395 return ret; 1651 return ret;
1396} 1652}
1397 1653
1654static int find_shared_port(struct file *fp,
1655 const struct ipath_user_info *uinfo)
1656{
1657 int devmax, ndev, i;
1658 int ret = 0;
1659
1660 devmax = ipath_count_units(NULL, NULL, NULL);
1661
1662 for (ndev = 0; ndev < devmax; ndev++) {
1663 struct ipath_devdata *dd = ipath_lookup(ndev);
1664
1665 if (!dd)
1666 continue;
1667 for (i = 1; i < dd->ipath_cfgports; i++) {
1668 struct ipath_portdata *pd = dd->ipath_pd[i];
1669
1670 /* Skip ports which are not yet open */
1671 if (!pd || !pd->port_cnt)
1672 continue;
1673 /* Skip port if it doesn't match the requested one */
1674 if (pd->port_subport_id != uinfo->spu_subport_id)
1675 continue;
1676 /* Verify the sharing process matches the master */
1677 if (pd->port_subport_cnt != uinfo->spu_subport_cnt ||
1678 pd->userversion != uinfo->spu_userversion ||
1679 pd->port_cnt >= pd->port_subport_cnt) {
1680 ret = -EINVAL;
1681 goto done;
1682 }
1683 port_fp(fp) = pd;
1684 subport_fp(fp) = pd->port_cnt++;
1685 tidcursor_fp(fp) = 0;
1686 pd->active_slaves |= 1 << subport_fp(fp);
1687 ipath_cdbg(PROC,
1688 "%s[%u] %u sharing %s[%u] unit:port %u:%u\n",
1689 current->comm, current->pid,
1690 subport_fp(fp),
1691 pd->port_comm, pd->port_pid,
1692 dd->ipath_unit, pd->port_port);
1693 ret = 1;
1694 goto done;
1695 }
1696 }
1697
1698done:
1699 return ret;
1700}
1701
1398static int ipath_open(struct inode *in, struct file *fp) 1702static int ipath_open(struct inode *in, struct file *fp)
1399{ 1703{
1400 int ret, user_minor; 1704 /* The real work is performed later in ipath_assign_port() */
1705 fp->private_data = kzalloc(sizeof(struct ipath_filedata), GFP_KERNEL);
1706 return fp->private_data ? 0 : -ENOMEM;
1707}
1708
1709
1710/* Get port early, so can set affinity prior to memory allocation */
1711static int ipath_assign_port(struct file *fp,
1712 const struct ipath_user_info *uinfo)
1713{
1714 int ret;
1715 int i_minor;
1716 unsigned swminor;
1717
1718 /* Check to be sure we haven't already initialized this file */
1719 if (port_fp(fp)) {
1720 ret = -EINVAL;
1721 goto done;
1722 }
1723
1724 /* for now, if major version is different, bail */
1725 if ((uinfo->spu_userversion >> 16) != IPATH_USER_SWMAJOR) {
1726 ipath_dbg("User major version %d not same as driver "
1727 "major %d\n", uinfo->spu_userversion >> 16,
1728 IPATH_USER_SWMAJOR);
1729 ret = -ENODEV;
1730 goto done;
1731 }
1732
1733 swminor = uinfo->spu_userversion & 0xffff;
1734 if (swminor != IPATH_USER_SWMINOR)
1735 ipath_dbg("User minor version %d not same as driver "
1736 "minor %d\n", swminor, IPATH_USER_SWMINOR);
1401 1737
1402 mutex_lock(&ipath_mutex); 1738 mutex_lock(&ipath_mutex);
1403 1739
1404 user_minor = iminor(in) - IPATH_USER_MINOR_BASE; 1740 if (swminor == IPATH_USER_SWMINOR && uinfo->spu_subport_cnt &&
1741 (ret = find_shared_port(fp, uinfo))) {
1742 mutex_unlock(&ipath_mutex);
1743 if (ret > 0)
1744 ret = 0;
1745 goto done;
1746 }
1747
1748 i_minor = iminor(fp->f_dentry->d_inode) - IPATH_USER_MINOR_BASE;
1405 ipath_cdbg(VERBOSE, "open on dev %lx (minor %d)\n", 1749 ipath_cdbg(VERBOSE, "open on dev %lx (minor %d)\n",
1406 (long)in->i_rdev, user_minor); 1750 (long)fp->f_dentry->d_inode->i_rdev, i_minor);
1407 1751
1408 if (user_minor) 1752 if (i_minor)
1409 ret = find_free_port(user_minor - 1, fp); 1753 ret = find_free_port(i_minor - 1, fp, uinfo);
1410 else 1754 else
1411 ret = find_best_unit(fp); 1755 ret = find_best_unit(fp, uinfo);
1412 1756
1413 mutex_unlock(&ipath_mutex); 1757 mutex_unlock(&ipath_mutex);
1758
1759done:
1760 return ret;
1761}
1762
1763
1764static int ipath_do_user_init(struct file *fp,
1765 const struct ipath_user_info *uinfo)
1766{
1767 int ret;
1768 struct ipath_portdata *pd;
1769 struct ipath_devdata *dd;
1770 u32 head32;
1771
1772 pd = port_fp(fp);
1773 dd = pd->port_dd;
1774
1775 if (uinfo->spu_rcvhdrsize) {
1776 ret = ipath_setrcvhdrsize(dd, uinfo->spu_rcvhdrsize);
1777 if (ret)
1778 goto done;
1779 }
1780
1781 /* for now we do nothing with rcvhdrcnt: uinfo->spu_rcvhdrcnt */
1782
1783 /* for right now, kernel piobufs are at end, so port 1 is at 0 */
1784 pd->port_piobufs = dd->ipath_piobufbase +
1785 dd->ipath_pbufsport * (pd->port_port - 1) * dd->ipath_palign;
1786 ipath_cdbg(VERBOSE, "Set base of piobufs for port %u to 0x%x\n",
1787 pd->port_port, pd->port_piobufs);
1788
1789 /*
1790 * Now allocate the rcvhdr Q and eager TIDs; skip the TID
1791 * array for time being. If pd->port_port > chip-supported,
1792 * we need to do extra stuff here to handle by handling overflow
1793 * through port 0, someday
1794 */
1795 ret = ipath_create_rcvhdrq(dd, pd);
1796 if (!ret)
1797 ret = ipath_create_user_egr(pd);
1798 if (ret)
1799 goto done;
1800
1801 /*
1802 * set the eager head register for this port to the current values
1803 * of the tail pointers, since we don't know if they were
1804 * updated on last use of the port.
1805 */
1806 head32 = ipath_read_ureg32(dd, ur_rcvegrindextail, pd->port_port);
1807 ipath_write_ureg(dd, ur_rcvegrindexhead, head32, pd->port_port);
1808 dd->ipath_lastegrheads[pd->port_port] = -1;
1809 dd->ipath_lastrcvhdrqtails[pd->port_port] = -1;
1810 ipath_cdbg(VERBOSE, "Wrote port%d egrhead %x from tail regs\n",
1811 pd->port_port, head32);
1812 pd->port_tidcursor = 0; /* start at beginning after open */
1813 /*
1814 * now enable the port; the tail registers will be written to memory
1815 * by the chip as soon as it sees the write to
1816 * dd->ipath_kregs->kr_rcvctrl. The update only happens on
1817 * transition from 0 to 1, so clear it first, then set it as part of
1818 * enabling the port. This will (very briefly) affect any other
1819 * open ports, but it shouldn't be long enough to be an issue.
1820 * We explictly set the in-memory copy to 0 beforehand, so we don't
1821 * have to wait to be sure the DMA update has happened.
1822 */
1823 *(volatile u64 *)pd->port_rcvhdrtail_kvaddr = 0ULL;
1824 set_bit(INFINIPATH_R_PORTENABLE_SHIFT + pd->port_port,
1825 &dd->ipath_rcvctrl);
1826 ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
1827 dd->ipath_rcvctrl & ~INFINIPATH_R_TAILUPD);
1828 ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
1829 dd->ipath_rcvctrl);
1830done:
1414 return ret; 1831 return ret;
1415} 1832}
1416 1833
@@ -1433,6 +1850,8 @@ static void unlock_expected_tids(struct ipath_portdata *pd)
1433 if (!dd->ipath_pageshadow[i]) 1850 if (!dd->ipath_pageshadow[i])
1434 continue; 1851 continue;
1435 1852
1853 pci_unmap_page(dd->pcidev, dd->ipath_physshadow[i],
1854 PAGE_SIZE, PCI_DMA_FROMDEVICE);
1436 ipath_release_user_pages_on_close(&dd->ipath_pageshadow[i], 1855 ipath_release_user_pages_on_close(&dd->ipath_pageshadow[i],
1437 1); 1856 1);
1438 dd->ipath_pageshadow[i] = NULL; 1857 dd->ipath_pageshadow[i] = NULL;
@@ -1453,6 +1872,7 @@ static void unlock_expected_tids(struct ipath_portdata *pd)
1453static int ipath_close(struct inode *in, struct file *fp) 1872static int ipath_close(struct inode *in, struct file *fp)
1454{ 1873{
1455 int ret = 0; 1874 int ret = 0;
1875 struct ipath_filedata *fd;
1456 struct ipath_portdata *pd; 1876 struct ipath_portdata *pd;
1457 struct ipath_devdata *dd; 1877 struct ipath_devdata *dd;
1458 unsigned port; 1878 unsigned port;
@@ -1462,9 +1882,24 @@ static int ipath_close(struct inode *in, struct file *fp)
1462 1882
1463 mutex_lock(&ipath_mutex); 1883 mutex_lock(&ipath_mutex);
1464 1884
1465 pd = port_fp(fp); 1885 fd = (struct ipath_filedata *) fp->private_data;
1466 port = pd->port_port;
1467 fp->private_data = NULL; 1886 fp->private_data = NULL;
1887 pd = fd->pd;
1888 if (!pd) {
1889 mutex_unlock(&ipath_mutex);
1890 goto bail;
1891 }
1892 if (--pd->port_cnt) {
1893 /*
1894 * XXX If the master closes the port before the slave(s),
1895 * revoke the mmap for the eager receive queue so
1896 * the slave(s) don't wait for receive data forever.
1897 */
1898 pd->active_slaves &= ~(1 << fd->subport);
1899 mutex_unlock(&ipath_mutex);
1900 goto bail;
1901 }
1902 port = pd->port_port;
1468 dd = pd->port_dd; 1903 dd = pd->port_dd;
1469 1904
1470 if (pd->port_hdrqfull) { 1905 if (pd->port_hdrqfull) {
@@ -1503,8 +1938,6 @@ static int ipath_close(struct inode *in, struct file *fp)
1503 1938
1504 /* clean up the pkeys for this port user */ 1939 /* clean up the pkeys for this port user */
1505 ipath_clean_part_key(pd, dd); 1940 ipath_clean_part_key(pd, dd);
1506
1507
1508 /* 1941 /*
1509 * be paranoid, and never write 0's to these, just use an 1942 * be paranoid, and never write 0's to these, just use an
1510 * unused part of the port 0 tail page. Of course, 1943 * unused part of the port 0 tail page. Of course,
@@ -1523,39 +1956,49 @@ static int ipath_close(struct inode *in, struct file *fp)
1523 i = dd->ipath_pbufsport * (port - 1); 1956 i = dd->ipath_pbufsport * (port - 1);
1524 ipath_disarm_piobufs(dd, i, dd->ipath_pbufsport); 1957 ipath_disarm_piobufs(dd, i, dd->ipath_pbufsport);
1525 1958
1959 dd->ipath_f_clear_tids(dd, pd->port_port);
1960
1526 if (dd->ipath_pageshadow) 1961 if (dd->ipath_pageshadow)
1527 unlock_expected_tids(pd); 1962 unlock_expected_tids(pd);
1528 ipath_stats.sps_ports--; 1963 ipath_stats.sps_ports--;
1529 ipath_cdbg(PROC, "%s[%u] closed port %u:%u\n", 1964 ipath_cdbg(PROC, "%s[%u] closed port %u:%u\n",
1530 pd->port_comm, pd->port_pid, 1965 pd->port_comm, pd->port_pid,
1531 dd->ipath_unit, port); 1966 dd->ipath_unit, port);
1532
1533 dd->ipath_f_clear_tids(dd, pd->port_port);
1534 } 1967 }
1535 1968
1536 pd->port_cnt = 0;
1537 pd->port_pid = 0; 1969 pd->port_pid = 0;
1538
1539 dd->ipath_pd[pd->port_port] = NULL; /* before releasing mutex */ 1970 dd->ipath_pd[pd->port_port] = NULL; /* before releasing mutex */
1540 mutex_unlock(&ipath_mutex); 1971 mutex_unlock(&ipath_mutex);
1541 ipath_free_pddata(dd, pd); /* after releasing the mutex */ 1972 ipath_free_pddata(dd, pd); /* after releasing the mutex */
1542 1973
1974bail:
1975 kfree(fd);
1543 return ret; 1976 return ret;
1544} 1977}
1545 1978
1546static int ipath_port_info(struct ipath_portdata *pd, 1979static int ipath_port_info(struct ipath_portdata *pd, u16 subport,
1547 struct ipath_port_info __user *uinfo) 1980 struct ipath_port_info __user *uinfo)
1548{ 1981{
1549 struct ipath_port_info info; 1982 struct ipath_port_info info;
1550 int nup; 1983 int nup;
1551 int ret; 1984 int ret;
1985 size_t sz;
1552 1986
1553 (void) ipath_count_units(NULL, &nup, NULL); 1987 (void) ipath_count_units(NULL, &nup, NULL);
1554 info.num_active = nup; 1988 info.num_active = nup;
1555 info.unit = pd->port_dd->ipath_unit; 1989 info.unit = pd->port_dd->ipath_unit;
1556 info.port = pd->port_port; 1990 info.port = pd->port_port;
1991 info.subport = subport;
1992 /* Don't return new fields if old library opened the port. */
1993 if ((pd->userversion & 0xffff) == IPATH_USER_SWMINOR) {
1994 /* Number of user ports available for this device. */
1995 info.num_ports = pd->port_dd->ipath_cfgports - 1;
1996 info.num_subports = pd->port_subport_cnt;
1997 sz = sizeof(info);
1998 } else
1999 sz = sizeof(info) - 2 * sizeof(u16);
1557 2000
1558 if (copy_to_user(uinfo, &info, sizeof(info))) { 2001 if (copy_to_user(uinfo, &info, sz)) {
1559 ret = -EFAULT; 2002 ret = -EFAULT;
1560 goto bail; 2003 goto bail;
1561 } 2004 }
@@ -1565,6 +2008,16 @@ bail:
1565 return ret; 2008 return ret;
1566} 2009}
1567 2010
2011static int ipath_get_slave_info(struct ipath_portdata *pd,
2012 void __user *slave_mask_addr)
2013{
2014 int ret = 0;
2015
2016 if (copy_to_user(slave_mask_addr, &pd->active_slaves, sizeof(u32)))
2017 ret = -EFAULT;
2018 return ret;
2019}
2020
1568static ssize_t ipath_write(struct file *fp, const char __user *data, 2021static ssize_t ipath_write(struct file *fp, const char __user *data,
1569 size_t count, loff_t *off) 2022 size_t count, loff_t *off)
1570{ 2023{
@@ -1591,6 +2044,8 @@ static ssize_t ipath_write(struct file *fp, const char __user *data,
1591 consumed = sizeof(cmd.type); 2044 consumed = sizeof(cmd.type);
1592 2045
1593 switch (cmd.type) { 2046 switch (cmd.type) {
2047 case IPATH_CMD_ASSIGN_PORT:
2048 case __IPATH_CMD_USER_INIT:
1594 case IPATH_CMD_USER_INIT: 2049 case IPATH_CMD_USER_INIT:
1595 copy = sizeof(cmd.cmd.user_info); 2050 copy = sizeof(cmd.cmd.user_info);
1596 dest = &cmd.cmd.user_info; 2051 dest = &cmd.cmd.user_info;
@@ -1617,6 +2072,11 @@ static ssize_t ipath_write(struct file *fp, const char __user *data,
1617 dest = &cmd.cmd.part_key; 2072 dest = &cmd.cmd.part_key;
1618 src = &ucmd->cmd.part_key; 2073 src = &ucmd->cmd.part_key;
1619 break; 2074 break;
2075 case IPATH_CMD_SLAVE_INFO:
2076 copy = sizeof(cmd.cmd.slave_mask_addr);
2077 dest = &cmd.cmd.slave_mask_addr;
2078 src = &ucmd->cmd.slave_mask_addr;
2079 break;
1620 default: 2080 default:
1621 ret = -EINVAL; 2081 ret = -EINVAL;
1622 goto bail; 2082 goto bail;
@@ -1634,34 +2094,55 @@ static ssize_t ipath_write(struct file *fp, const char __user *data,
1634 2094
1635 consumed += copy; 2095 consumed += copy;
1636 pd = port_fp(fp); 2096 pd = port_fp(fp);
2097 if (!pd && cmd.type != __IPATH_CMD_USER_INIT &&
2098 cmd.type != IPATH_CMD_ASSIGN_PORT) {
2099 ret = -EINVAL;
2100 goto bail;
2101 }
1637 2102
1638 switch (cmd.type) { 2103 switch (cmd.type) {
2104 case IPATH_CMD_ASSIGN_PORT:
2105 ret = ipath_assign_port(fp, &cmd.cmd.user_info);
2106 if (ret)
2107 goto bail;
2108 break;
2109 case __IPATH_CMD_USER_INIT:
2110 /* backwards compatibility, get port first */
2111 ret = ipath_assign_port(fp, &cmd.cmd.user_info);
2112 if (ret)
2113 goto bail;
2114 /* and fall through to current version. */
1639 case IPATH_CMD_USER_INIT: 2115 case IPATH_CMD_USER_INIT:
1640 ret = ipath_do_user_init(pd, &cmd.cmd.user_info); 2116 ret = ipath_do_user_init(fp, &cmd.cmd.user_info);
1641 if (ret < 0) 2117 if (ret)
1642 goto bail; 2118 goto bail;
1643 ret = ipath_get_base_info( 2119 ret = ipath_get_base_info(
1644 pd, (void __user *) (unsigned long) 2120 fp, (void __user *) (unsigned long)
1645 cmd.cmd.user_info.spu_base_info, 2121 cmd.cmd.user_info.spu_base_info,
1646 cmd.cmd.user_info.spu_base_info_size); 2122 cmd.cmd.user_info.spu_base_info_size);
1647 break; 2123 break;
1648 case IPATH_CMD_RECV_CTRL: 2124 case IPATH_CMD_RECV_CTRL:
1649 ret = ipath_manage_rcvq(pd, cmd.cmd.recv_ctrl); 2125 ret = ipath_manage_rcvq(pd, subport_fp(fp), cmd.cmd.recv_ctrl);
1650 break; 2126 break;
1651 case IPATH_CMD_PORT_INFO: 2127 case IPATH_CMD_PORT_INFO:
1652 ret = ipath_port_info(pd, 2128 ret = ipath_port_info(pd, subport_fp(fp),
1653 (struct ipath_port_info __user *) 2129 (struct ipath_port_info __user *)
1654 (unsigned long) cmd.cmd.port_info); 2130 (unsigned long) cmd.cmd.port_info);
1655 break; 2131 break;
1656 case IPATH_CMD_TID_UPDATE: 2132 case IPATH_CMD_TID_UPDATE:
1657 ret = ipath_tid_update(pd, &cmd.cmd.tid_info); 2133 ret = ipath_tid_update(pd, fp, &cmd.cmd.tid_info);
1658 break; 2134 break;
1659 case IPATH_CMD_TID_FREE: 2135 case IPATH_CMD_TID_FREE:
1660 ret = ipath_tid_free(pd, &cmd.cmd.tid_info); 2136 ret = ipath_tid_free(pd, subport_fp(fp), &cmd.cmd.tid_info);
1661 break; 2137 break;
1662 case IPATH_CMD_SET_PART_KEY: 2138 case IPATH_CMD_SET_PART_KEY:
1663 ret = ipath_set_part_key(pd, cmd.cmd.part_key); 2139 ret = ipath_set_part_key(pd, cmd.cmd.part_key);
1664 break; 2140 break;
2141 case IPATH_CMD_SLAVE_INFO:
2142 ret = ipath_get_slave_info(pd,
2143 (void __user *) (unsigned long)
2144 cmd.cmd.slave_mask_addr);
2145 break;
1665 } 2146 }
1666 2147
1667 if (ret >= 0) 2148 if (ret >= 0)
@@ -1858,4 +2339,3 @@ void ipath_user_remove(struct ipath_devdata *dd)
1858bail: 2339bail:
1859 return; 2340 return;
1860} 2341}
1861
diff --git a/drivers/infiniband/hw/ipath/ipath_fs.c b/drivers/infiniband/hw/ipath/ipath_fs.c
index c8a8af0fe471..a507d0b5be6c 100644
--- a/drivers/infiniband/hw/ipath/ipath_fs.c
+++ b/drivers/infiniband/hw/ipath/ipath_fs.c
@@ -356,19 +356,16 @@ static ssize_t flash_write(struct file *file, const char __user *buf,
356 356
357 pos = *ppos; 357 pos = *ppos;
358 358
359 if ( pos < 0) { 359 if (pos != 0) {
360 ret = -EINVAL; 360 ret = -EINVAL;
361 goto bail; 361 goto bail;
362 } 362 }
363 363
364 if (pos >= sizeof(struct ipath_flash)) { 364 if (count != sizeof(struct ipath_flash)) {
365 ret = 0; 365 ret = -EINVAL;
366 goto bail; 366 goto bail;
367 } 367 }
368 368
369 if (count > sizeof(struct ipath_flash) - pos)
370 count = sizeof(struct ipath_flash) - pos;
371
372 tmp = kmalloc(count, GFP_KERNEL); 369 tmp = kmalloc(count, GFP_KERNEL);
373 if (!tmp) { 370 if (!tmp) {
374 ret = -ENOMEM; 371 ret = -ENOMEM;
diff --git a/drivers/infiniband/hw/ipath/ipath_iba6110.c b/drivers/infiniband/hw/ipath/ipath_iba6110.c
index 5c9b509e40e4..9e4e8d4c6e20 100644
--- a/drivers/infiniband/hw/ipath/ipath_iba6110.c
+++ b/drivers/infiniband/hw/ipath/ipath_iba6110.c
@@ -252,8 +252,8 @@ static const struct ipath_cregs ipath_ht_cregs = {
252}; 252};
253 253
254/* kr_intstatus, kr_intclear, kr_intmask bits */ 254/* kr_intstatus, kr_intclear, kr_intmask bits */
255#define INFINIPATH_I_RCVURG_MASK 0x1FF 255#define INFINIPATH_I_RCVURG_MASK ((1U<<9)-1)
256#define INFINIPATH_I_RCVAVAIL_MASK 0x1FF 256#define INFINIPATH_I_RCVAVAIL_MASK ((1U<<9)-1)
257 257
258/* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */ 258/* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */
259#define INFINIPATH_HWE_HTCMEMPARITYERR_SHIFT 0 259#define INFINIPATH_HWE_HTCMEMPARITYERR_SHIFT 0
@@ -338,7 +338,7 @@ static void hwerr_crcbits(struct ipath_devdata *dd, ipath_err_t hwerrs,
338 if (crcbits) { 338 if (crcbits) {
339 u16 ctrl0, ctrl1; 339 u16 ctrl0, ctrl1;
340 snprintf(bitsmsg, sizeof bitsmsg, 340 snprintf(bitsmsg, sizeof bitsmsg,
341 "[HT%s lane %s CRC (%llx); ignore till reload]", 341 "[HT%s lane %s CRC (%llx); powercycle to completely clear]",
342 !(crcbits & _IPATH_HTLINK1_CRCBITS) ? 342 !(crcbits & _IPATH_HTLINK1_CRCBITS) ?
343 "0 (A)" : (!(crcbits & _IPATH_HTLINK0_CRCBITS) 343 "0 (A)" : (!(crcbits & _IPATH_HTLINK0_CRCBITS)
344 ? "1 (B)" : "0+1 (A+B)"), 344 ? "1 (B)" : "0+1 (A+B)"),
@@ -389,17 +389,28 @@ static void hwerr_crcbits(struct ipath_devdata *dd, ipath_err_t hwerrs,
389 _IPATH_HTLINK1_CRCBITS))); 389 _IPATH_HTLINK1_CRCBITS)));
390} 390}
391 391
392/* 6110 specific hardware errors... */
393static const struct ipath_hwerror_msgs ipath_6110_hwerror_msgs[] = {
394 INFINIPATH_HWE_MSG(HTCBUSIREQPARITYERR, "HTC Ireq Parity"),
395 INFINIPATH_HWE_MSG(HTCBUSTREQPARITYERR, "HTC Treq Parity"),
396 INFINIPATH_HWE_MSG(HTCBUSTRESPPARITYERR, "HTC Tresp Parity"),
397 INFINIPATH_HWE_MSG(HTCMISCERR5, "HT core Misc5"),
398 INFINIPATH_HWE_MSG(HTCMISCERR6, "HT core Misc6"),
399 INFINIPATH_HWE_MSG(HTCMISCERR7, "HT core Misc7"),
400 INFINIPATH_HWE_MSG(RXDSYNCMEMPARITYERR, "Rx Dsync"),
401 INFINIPATH_HWE_MSG(SERDESPLLFAILED, "SerDes PLL"),
402};
403
392/** 404/**
393 * ipath_ht_handle_hwerrors - display hardware errors 405 * ipath_ht_handle_hwerrors - display hardware errors.
394 * @dd: the infinipath device 406 * @dd: the infinipath device
395 * @msg: the output buffer 407 * @msg: the output buffer
396 * @msgl: the size of the output buffer 408 * @msgl: the size of the output buffer
397 * 409 *
398 * Use same msg buffer as regular errors to avoid 410 * Use same msg buffer as regular errors to avoid excessive stack
399 * excessive stack use. Most hardware errors are catastrophic, but for 411 * use. Most hardware errors are catastrophic, but for right now,
400 * right now, we'll print them and continue. 412 * we'll print them and continue. We reuse the same message buffer as
401 * We reuse the same message buffer as ipath_handle_errors() to avoid 413 * ipath_handle_errors() to avoid excessive stack usage.
402 * excessive stack usage.
403 */ 414 */
404static void ipath_ht_handle_hwerrors(struct ipath_devdata *dd, char *msg, 415static void ipath_ht_handle_hwerrors(struct ipath_devdata *dd, char *msg,
405 size_t msgl) 416 size_t msgl)
@@ -440,19 +451,49 @@ static void ipath_ht_handle_hwerrors(struct ipath_devdata *dd, char *msg,
440 * make sure we get this much out, unless told to be quiet, 451 * make sure we get this much out, unless told to be quiet,
441 * or it's occurred within the last 5 seconds 452 * or it's occurred within the last 5 seconds
442 */ 453 */
443 if ((hwerrs & ~dd->ipath_lasthwerror) || 454 if ((hwerrs & ~(dd->ipath_lasthwerror |
455 ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF |
456 INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC)
457 << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT))) ||
444 (ipath_debug & __IPATH_VERBDBG)) 458 (ipath_debug & __IPATH_VERBDBG))
445 dev_info(&dd->pcidev->dev, "Hardware error: hwerr=0x%llx " 459 dev_info(&dd->pcidev->dev, "Hardware error: hwerr=0x%llx "
446 "(cleared)\n", (unsigned long long) hwerrs); 460 "(cleared)\n", (unsigned long long) hwerrs);
447 dd->ipath_lasthwerror |= hwerrs; 461 dd->ipath_lasthwerror |= hwerrs;
448 462
449 if (hwerrs & ~infinipath_hwe_bitsextant) 463 if (hwerrs & ~dd->ipath_hwe_bitsextant)
450 ipath_dev_err(dd, "hwerror interrupt with unknown errors " 464 ipath_dev_err(dd, "hwerror interrupt with unknown errors "
451 "%llx set\n", (unsigned long long) 465 "%llx set\n", (unsigned long long)
452 (hwerrs & ~infinipath_hwe_bitsextant)); 466 (hwerrs & ~dd->ipath_hwe_bitsextant));
453 467
454 ctrl = ipath_read_kreg32(dd, dd->ipath_kregs->kr_control); 468 ctrl = ipath_read_kreg32(dd, dd->ipath_kregs->kr_control);
455 if (ctrl & INFINIPATH_C_FREEZEMODE) { 469 if (ctrl & INFINIPATH_C_FREEZEMODE) {
470 /*
471 * parity errors in send memory are recoverable,
472 * just cancel the send (if indicated in * sendbuffererror),
473 * count the occurrence, unfreeze (if no other handled
474 * hardware error bits are set), and continue. They can
475 * occur if a processor speculative read is done to the PIO
476 * buffer while we are sending a packet, for example.
477 */
478 if (hwerrs & ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF |
479 INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC)
480 << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT)) {
481 ipath_stats.sps_txeparity++;
482 ipath_dbg("Recovering from TXE parity error (%llu), "
483 "hwerrstatus=%llx\n",
484 (unsigned long long) ipath_stats.sps_txeparity,
485 (unsigned long long) hwerrs);
486 ipath_disarm_senderrbufs(dd);
487 hwerrs &= ~((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF |
488 INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC)
489 << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT);
490 if (!hwerrs) { /* else leave in freeze mode */
491 ipath_write_kreg(dd,
492 dd->ipath_kregs->kr_control,
493 dd->ipath_control);
494 return;
495 }
496 }
456 if (hwerrs) { 497 if (hwerrs) {
457 /* 498 /*
458 * if any set that we aren't ignoring; only 499 * if any set that we aren't ignoring; only
@@ -499,44 +540,16 @@ static void ipath_ht_handle_hwerrors(struct ipath_devdata *dd, char *msg,
499 bits); 540 bits);
500 strlcat(msg, bitsmsg, msgl); 541 strlcat(msg, bitsmsg, msgl);
501 } 542 }
502 if (hwerrs & (INFINIPATH_HWE_RXEMEMPARITYERR_MASK 543
503 << INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT)) { 544 ipath_format_hwerrors(hwerrs,
504 bits = (u32) ((hwerrs >> 545 ipath_6110_hwerror_msgs,
505 INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT) & 546 sizeof(ipath_6110_hwerror_msgs) /
506 INFINIPATH_HWE_RXEMEMPARITYERR_MASK); 547 sizeof(ipath_6110_hwerror_msgs[0]),
507 snprintf(bitsmsg, sizeof bitsmsg, "[RXE Parity Errs %x] ", 548 msg, msgl);
508 bits);
509 strlcat(msg, bitsmsg, msgl);
510 }
511 if (hwerrs & (INFINIPATH_HWE_TXEMEMPARITYERR_MASK
512 << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT)) {
513 bits = (u32) ((hwerrs >>
514 INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT) &
515 INFINIPATH_HWE_TXEMEMPARITYERR_MASK);
516 snprintf(bitsmsg, sizeof bitsmsg, "[TXE Parity Errs %x] ",
517 bits);
518 strlcat(msg, bitsmsg, msgl);
519 }
520 if (hwerrs & INFINIPATH_HWE_IBCBUSTOSPCPARITYERR)
521 strlcat(msg, "[IB2IPATH Parity]", msgl);
522 if (hwerrs & INFINIPATH_HWE_IBCBUSFRSPCPARITYERR)
523 strlcat(msg, "[IPATH2IB Parity]", msgl);
524 if (hwerrs & INFINIPATH_HWE_HTCBUSIREQPARITYERR)
525 strlcat(msg, "[HTC Ireq Parity]", msgl);
526 if (hwerrs & INFINIPATH_HWE_HTCBUSTREQPARITYERR)
527 strlcat(msg, "[HTC Treq Parity]", msgl);
528 if (hwerrs & INFINIPATH_HWE_HTCBUSTRESPPARITYERR)
529 strlcat(msg, "[HTC Tresp Parity]", msgl);
530 549
531 if (hwerrs & (_IPATH_HTLINK0_CRCBITS | _IPATH_HTLINK1_CRCBITS)) 550 if (hwerrs & (_IPATH_HTLINK0_CRCBITS | _IPATH_HTLINK1_CRCBITS))
532 hwerr_crcbits(dd, hwerrs, msg, msgl); 551 hwerr_crcbits(dd, hwerrs, msg, msgl);
533 552
534 if (hwerrs & INFINIPATH_HWE_HTCMISCERR5)
535 strlcat(msg, "[HT core Misc5]", msgl);
536 if (hwerrs & INFINIPATH_HWE_HTCMISCERR6)
537 strlcat(msg, "[HT core Misc6]", msgl);
538 if (hwerrs & INFINIPATH_HWE_HTCMISCERR7)
539 strlcat(msg, "[HT core Misc7]", msgl);
540 if (hwerrs & INFINIPATH_HWE_MEMBISTFAILED) { 553 if (hwerrs & INFINIPATH_HWE_MEMBISTFAILED) {
541 strlcat(msg, "[Memory BIST test failed, InfiniPath hardware unusable]", 554 strlcat(msg, "[Memory BIST test failed, InfiniPath hardware unusable]",
542 msgl); 555 msgl);
@@ -573,11 +586,6 @@ static void ipath_ht_handle_hwerrors(struct ipath_devdata *dd, char *msg,
573 dd->ipath_hwerrmask); 586 dd->ipath_hwerrmask);
574 } 587 }
575 588
576 if (hwerrs & INFINIPATH_HWE_RXDSYNCMEMPARITYERR)
577 strlcat(msg, "[Rx Dsync]", msgl);
578 if (hwerrs & INFINIPATH_HWE_SERDESPLLFAILED)
579 strlcat(msg, "[SerDes PLL]", msgl);
580
581 ipath_dev_err(dd, "%s hardware error\n", msg); 589 ipath_dev_err(dd, "%s hardware error\n", msg);
582 if (isfatal && !ipath_diag_inuse && dd->ipath_freezemsg) 590 if (isfatal && !ipath_diag_inuse && dd->ipath_freezemsg)
583 /* 591 /*
@@ -1080,21 +1088,21 @@ static void ipath_setup_ht_setextled(struct ipath_devdata *dd,
1080 ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, extctl); 1088 ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, extctl);
1081} 1089}
1082 1090
1083static void ipath_init_ht_variables(void) 1091static void ipath_init_ht_variables(struct ipath_devdata *dd)
1084{ 1092{
1085 ipath_gpio_sda_num = _IPATH_GPIO_SDA_NUM; 1093 dd->ipath_gpio_sda_num = _IPATH_GPIO_SDA_NUM;
1086 ipath_gpio_scl_num = _IPATH_GPIO_SCL_NUM; 1094 dd->ipath_gpio_scl_num = _IPATH_GPIO_SCL_NUM;
1087 ipath_gpio_sda = IPATH_GPIO_SDA; 1095 dd->ipath_gpio_sda = IPATH_GPIO_SDA;
1088 ipath_gpio_scl = IPATH_GPIO_SCL; 1096 dd->ipath_gpio_scl = IPATH_GPIO_SCL;
1089 1097
1090 infinipath_i_bitsextant = 1098 dd->ipath_i_bitsextant =
1091 (INFINIPATH_I_RCVURG_MASK << INFINIPATH_I_RCVURG_SHIFT) | 1099 (INFINIPATH_I_RCVURG_MASK << INFINIPATH_I_RCVURG_SHIFT) |
1092 (INFINIPATH_I_RCVAVAIL_MASK << 1100 (INFINIPATH_I_RCVAVAIL_MASK <<
1093 INFINIPATH_I_RCVAVAIL_SHIFT) | 1101 INFINIPATH_I_RCVAVAIL_SHIFT) |
1094 INFINIPATH_I_ERROR | INFINIPATH_I_SPIOSENT | 1102 INFINIPATH_I_ERROR | INFINIPATH_I_SPIOSENT |
1095 INFINIPATH_I_SPIOBUFAVAIL | INFINIPATH_I_GPIO; 1103 INFINIPATH_I_SPIOBUFAVAIL | INFINIPATH_I_GPIO;
1096 1104
1097 infinipath_e_bitsextant = 1105 dd->ipath_e_bitsextant =
1098 INFINIPATH_E_RFORMATERR | INFINIPATH_E_RVCRC | 1106 INFINIPATH_E_RFORMATERR | INFINIPATH_E_RVCRC |
1099 INFINIPATH_E_RICRC | INFINIPATH_E_RMINPKTLEN | 1107 INFINIPATH_E_RICRC | INFINIPATH_E_RMINPKTLEN |
1100 INFINIPATH_E_RMAXPKTLEN | INFINIPATH_E_RLONGPKTLEN | 1108 INFINIPATH_E_RMAXPKTLEN | INFINIPATH_E_RLONGPKTLEN |
@@ -1112,7 +1120,7 @@ static void ipath_init_ht_variables(void)
1112 INFINIPATH_E_INVALIDADDR | INFINIPATH_E_RESET | 1120 INFINIPATH_E_INVALIDADDR | INFINIPATH_E_RESET |
1113 INFINIPATH_E_HARDWARE; 1121 INFINIPATH_E_HARDWARE;
1114 1122
1115 infinipath_hwe_bitsextant = 1123 dd->ipath_hwe_bitsextant =
1116 (INFINIPATH_HWE_HTCMEMPARITYERR_MASK << 1124 (INFINIPATH_HWE_HTCMEMPARITYERR_MASK <<
1117 INFINIPATH_HWE_HTCMEMPARITYERR_SHIFT) | 1125 INFINIPATH_HWE_HTCMEMPARITYERR_SHIFT) |
1118 (INFINIPATH_HWE_TXEMEMPARITYERR_MASK << 1126 (INFINIPATH_HWE_TXEMEMPARITYERR_MASK <<
@@ -1141,8 +1149,8 @@ static void ipath_init_ht_variables(void)
1141 INFINIPATH_HWE_IBCBUSTOSPCPARITYERR | 1149 INFINIPATH_HWE_IBCBUSTOSPCPARITYERR |
1142 INFINIPATH_HWE_IBCBUSFRSPCPARITYERR; 1150 INFINIPATH_HWE_IBCBUSFRSPCPARITYERR;
1143 1151
1144 infinipath_i_rcvavail_mask = INFINIPATH_I_RCVAVAIL_MASK; 1152 dd->ipath_i_rcvavail_mask = INFINIPATH_I_RCVAVAIL_MASK;
1145 infinipath_i_rcvurg_mask = INFINIPATH_I_RCVURG_MASK; 1153 dd->ipath_i_rcvurg_mask = INFINIPATH_I_RCVURG_MASK;
1146} 1154}
1147 1155
1148/** 1156/**
@@ -1607,5 +1615,5 @@ void ipath_init_iba6110_funcs(struct ipath_devdata *dd)
1607 * do very early init that is needed before ipath_f_bus is 1615 * do very early init that is needed before ipath_f_bus is
1608 * called 1616 * called
1609 */ 1617 */
1610 ipath_init_ht_variables(); 1618 ipath_init_ht_variables(dd);
1611} 1619}
diff --git a/drivers/infiniband/hw/ipath/ipath_iba6120.c b/drivers/infiniband/hw/ipath/ipath_iba6120.c
index d86516d23df6..a72ab9de386a 100644
--- a/drivers/infiniband/hw/ipath/ipath_iba6120.c
+++ b/drivers/infiniband/hw/ipath/ipath_iba6120.c
@@ -263,8 +263,8 @@ static const struct ipath_cregs ipath_pe_cregs = {
263}; 263};
264 264
265/* kr_intstatus, kr_intclear, kr_intmask bits */ 265/* kr_intstatus, kr_intclear, kr_intmask bits */
266#define INFINIPATH_I_RCVURG_MASK 0x1F 266#define INFINIPATH_I_RCVURG_MASK ((1U<<5)-1)
267#define INFINIPATH_I_RCVAVAIL_MASK 0x1F 267#define INFINIPATH_I_RCVAVAIL_MASK ((1U<<5)-1)
268 268
269/* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */ 269/* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */
270#define INFINIPATH_HWE_PCIEMEMPARITYERR_MASK 0x000000000000003fULL 270#define INFINIPATH_HWE_PCIEMEMPARITYERR_MASK 0x000000000000003fULL
@@ -294,6 +294,33 @@ static const struct ipath_cregs ipath_pe_cregs = {
294#define IPATH_GPIO_SCL (1ULL << \ 294#define IPATH_GPIO_SCL (1ULL << \
295 (_IPATH_GPIO_SCL_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT)) 295 (_IPATH_GPIO_SCL_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT))
296 296
297/*
298 * Rev2 silicon allows suppressing check for ArmLaunch errors.
299 * this can speed up short packet sends on systems that do
300 * not guaranteee write-order.
301 */
302#define INFINIPATH_XGXS_SUPPRESS_ARMLAUNCH_ERR (1ULL<<63)
303
304/* 6120 specific hardware errors... */
305static const struct ipath_hwerror_msgs ipath_6120_hwerror_msgs[] = {
306 INFINIPATH_HWE_MSG(PCIEPOISONEDTLP, "PCIe Poisoned TLP"),
307 INFINIPATH_HWE_MSG(PCIECPLTIMEOUT, "PCIe completion timeout"),
308 /*
309 * In practice, it's unlikely wthat we'll see PCIe PLL, or bus
310 * parity or memory parity error failures, because most likely we
311 * won't be able to talk to the core of the chip. Nonetheless, we
312 * might see them, if they are in parts of the PCIe core that aren't
313 * essential.
314 */
315 INFINIPATH_HWE_MSG(PCIE1PLLFAILED, "PCIePLL1"),
316 INFINIPATH_HWE_MSG(PCIE0PLLFAILED, "PCIePLL0"),
317 INFINIPATH_HWE_MSG(PCIEBUSPARITYXTLH, "PCIe XTLH core parity"),
318 INFINIPATH_HWE_MSG(PCIEBUSPARITYXADM, "PCIe ADM TX core parity"),
319 INFINIPATH_HWE_MSG(PCIEBUSPARITYRADM, "PCIe ADM RX core parity"),
320 INFINIPATH_HWE_MSG(RXDSYNCMEMPARITYERR, "Rx Dsync"),
321 INFINIPATH_HWE_MSG(SERDESPLLFAILED, "SerDes PLL"),
322};
323
297/** 324/**
298 * ipath_pe_handle_hwerrors - display hardware errors. 325 * ipath_pe_handle_hwerrors - display hardware errors.
299 * @dd: the infinipath device 326 * @dd: the infinipath device
@@ -343,19 +370,49 @@ static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg,
343 * make sure we get this much out, unless told to be quiet, 370 * make sure we get this much out, unless told to be quiet,
344 * or it's occurred within the last 5 seconds 371 * or it's occurred within the last 5 seconds
345 */ 372 */
346 if ((hwerrs & ~dd->ipath_lasthwerror) || 373 if ((hwerrs & ~(dd->ipath_lasthwerror |
374 ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF |
375 INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC)
376 << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT))) ||
347 (ipath_debug & __IPATH_VERBDBG)) 377 (ipath_debug & __IPATH_VERBDBG))
348 dev_info(&dd->pcidev->dev, "Hardware error: hwerr=0x%llx " 378 dev_info(&dd->pcidev->dev, "Hardware error: hwerr=0x%llx "
349 "(cleared)\n", (unsigned long long) hwerrs); 379 "(cleared)\n", (unsigned long long) hwerrs);
350 dd->ipath_lasthwerror |= hwerrs; 380 dd->ipath_lasthwerror |= hwerrs;
351 381
352 if (hwerrs & ~infinipath_hwe_bitsextant) 382 if (hwerrs & ~dd->ipath_hwe_bitsextant)
353 ipath_dev_err(dd, "hwerror interrupt with unknown errors " 383 ipath_dev_err(dd, "hwerror interrupt with unknown errors "
354 "%llx set\n", (unsigned long long) 384 "%llx set\n", (unsigned long long)
355 (hwerrs & ~infinipath_hwe_bitsextant)); 385 (hwerrs & ~dd->ipath_hwe_bitsextant));
356 386
357 ctrl = ipath_read_kreg32(dd, dd->ipath_kregs->kr_control); 387 ctrl = ipath_read_kreg32(dd, dd->ipath_kregs->kr_control);
358 if (ctrl & INFINIPATH_C_FREEZEMODE) { 388 if (ctrl & INFINIPATH_C_FREEZEMODE) {
389 /*
390 * parity errors in send memory are recoverable,
391 * just cancel the send (if indicated in * sendbuffererror),
392 * count the occurrence, unfreeze (if no other handled
393 * hardware error bits are set), and continue. They can
394 * occur if a processor speculative read is done to the PIO
395 * buffer while we are sending a packet, for example.
396 */
397 if (hwerrs & ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF |
398 INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC)
399 << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT)) {
400 ipath_stats.sps_txeparity++;
401 ipath_dbg("Recovering from TXE parity error (%llu), "
402 "hwerrstatus=%llx\n",
403 (unsigned long long) ipath_stats.sps_txeparity,
404 (unsigned long long) hwerrs);
405 ipath_disarm_senderrbufs(dd);
406 hwerrs &= ~((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF |
407 INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC)
408 << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT);
409 if (!hwerrs) { /* else leave in freeze mode */
410 ipath_write_kreg(dd,
411 dd->ipath_kregs->kr_control,
412 dd->ipath_control);
413 return;
414 }
415 }
359 if (hwerrs) { 416 if (hwerrs) {
360 /* 417 /*
361 * if any set that we aren't ignoring only make the 418 * if any set that we aren't ignoring only make the
@@ -379,9 +436,8 @@ static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg,
379 } else { 436 } else {
380 ipath_dbg("Clearing freezemode on ignored hardware " 437 ipath_dbg("Clearing freezemode on ignored hardware "
381 "error\n"); 438 "error\n");
382 ctrl &= ~INFINIPATH_C_FREEZEMODE;
383 ipath_write_kreg(dd, dd->ipath_kregs->kr_control, 439 ipath_write_kreg(dd, dd->ipath_kregs->kr_control,
384 ctrl); 440 dd->ipath_control);
385 } 441 }
386 } 442 }
387 443
@@ -396,24 +452,13 @@ static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg,
396 ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask, 452 ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
397 dd->ipath_hwerrmask); 453 dd->ipath_hwerrmask);
398 } 454 }
399 if (hwerrs & (INFINIPATH_HWE_RXEMEMPARITYERR_MASK 455
400 << INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT)) { 456 ipath_format_hwerrors(hwerrs,
401 bits = (u32) ((hwerrs >> 457 ipath_6120_hwerror_msgs,
402 INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT) & 458 sizeof(ipath_6120_hwerror_msgs)/
403 INFINIPATH_HWE_RXEMEMPARITYERR_MASK); 459 sizeof(ipath_6120_hwerror_msgs[0]),
404 snprintf(bitsmsg, sizeof bitsmsg, "[RXE Parity Errs %x] ", 460 msg, msgl);
405 bits); 461
406 strlcat(msg, bitsmsg, msgl);
407 }
408 if (hwerrs & (INFINIPATH_HWE_TXEMEMPARITYERR_MASK
409 << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT)) {
410 bits = (u32) ((hwerrs >>
411 INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT) &
412 INFINIPATH_HWE_TXEMEMPARITYERR_MASK);
413 snprintf(bitsmsg, sizeof bitsmsg, "[TXE Parity Errs %x] ",
414 bits);
415 strlcat(msg, bitsmsg, msgl);
416 }
417 if (hwerrs & (INFINIPATH_HWE_PCIEMEMPARITYERR_MASK 462 if (hwerrs & (INFINIPATH_HWE_PCIEMEMPARITYERR_MASK
418 << INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT)) { 463 << INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT)) {
419 bits = (u32) ((hwerrs >> 464 bits = (u32) ((hwerrs >>
@@ -423,10 +468,6 @@ static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg,
423 "[PCIe Mem Parity Errs %x] ", bits); 468 "[PCIe Mem Parity Errs %x] ", bits);
424 strlcat(msg, bitsmsg, msgl); 469 strlcat(msg, bitsmsg, msgl);
425 } 470 }
426 if (hwerrs & INFINIPATH_HWE_IBCBUSTOSPCPARITYERR)
427 strlcat(msg, "[IB2IPATH Parity]", msgl);
428 if (hwerrs & INFINIPATH_HWE_IBCBUSFRSPCPARITYERR)
429 strlcat(msg, "[IPATH2IB Parity]", msgl);
430 471
431#define _IPATH_PLL_FAIL (INFINIPATH_HWE_COREPLL_FBSLIP | \ 472#define _IPATH_PLL_FAIL (INFINIPATH_HWE_COREPLL_FBSLIP | \
432 INFINIPATH_HWE_COREPLL_RFSLIP ) 473 INFINIPATH_HWE_COREPLL_RFSLIP )
@@ -452,34 +493,6 @@ static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg,
452 dd->ipath_hwerrmask); 493 dd->ipath_hwerrmask);
453 } 494 }
454 495
455 if (hwerrs & INFINIPATH_HWE_PCIEPOISONEDTLP)
456 strlcat(msg, "[PCIe Poisoned TLP]", msgl);
457 if (hwerrs & INFINIPATH_HWE_PCIECPLTIMEOUT)
458 strlcat(msg, "[PCIe completion timeout]", msgl);
459
460 /*
461 * In practice, it's unlikely wthat we'll see PCIe PLL, or bus
462 * parity or memory parity error failures, because most likely we
463 * won't be able to talk to the core of the chip. Nonetheless, we
464 * might see them, if they are in parts of the PCIe core that aren't
465 * essential.
466 */
467 if (hwerrs & INFINIPATH_HWE_PCIE1PLLFAILED)
468 strlcat(msg, "[PCIePLL1]", msgl);
469 if (hwerrs & INFINIPATH_HWE_PCIE0PLLFAILED)
470 strlcat(msg, "[PCIePLL0]", msgl);
471 if (hwerrs & INFINIPATH_HWE_PCIEBUSPARITYXTLH)
472 strlcat(msg, "[PCIe XTLH core parity]", msgl);
473 if (hwerrs & INFINIPATH_HWE_PCIEBUSPARITYXADM)
474 strlcat(msg, "[PCIe ADM TX core parity]", msgl);
475 if (hwerrs & INFINIPATH_HWE_PCIEBUSPARITYRADM)
476 strlcat(msg, "[PCIe ADM RX core parity]", msgl);
477
478 if (hwerrs & INFINIPATH_HWE_RXDSYNCMEMPARITYERR)
479 strlcat(msg, "[Rx Dsync]", msgl);
480 if (hwerrs & INFINIPATH_HWE_SERDESPLLFAILED)
481 strlcat(msg, "[SerDes PLL]", msgl);
482
483 ipath_dev_err(dd, "%s hardware error\n", msg); 496 ipath_dev_err(dd, "%s hardware error\n", msg);
484 if (isfatal && !ipath_diag_inuse && dd->ipath_freezemsg) { 497 if (isfatal && !ipath_diag_inuse && dd->ipath_freezemsg) {
485 /* 498 /*
@@ -525,6 +538,9 @@ static int ipath_pe_boardname(struct ipath_devdata *dd, char *name,
525 case 5: 538 case 5:
526 n = "InfiniPath_QMH7140"; 539 n = "InfiniPath_QMH7140";
527 break; 540 break;
541 case 6:
542 n = "InfiniPath_QLE7142";
543 break;
528 default: 544 default:
529 ipath_dev_err(dd, 545 ipath_dev_err(dd,
530 "Don't yet know about board with ID %u\n", 546 "Don't yet know about board with ID %u\n",
@@ -571,9 +587,12 @@ static void ipath_pe_init_hwerrors(struct ipath_devdata *dd)
571 if (!dd->ipath_boardrev) // no PLL for Emulator 587 if (!dd->ipath_boardrev) // no PLL for Emulator
572 val &= ~INFINIPATH_HWE_SERDESPLLFAILED; 588 val &= ~INFINIPATH_HWE_SERDESPLLFAILED;
573 589
574 /* workaround bug 9460 in internal interface bus parity checking */ 590 if (dd->ipath_minrev < 2) {
575 val &= ~INFINIPATH_HWE_PCIEBUSPARITYRADM; 591 /* workaround bug 9460 in internal interface bus parity
576 592 * checking. Fixed (HW bug 9490) in Rev2.
593 */
594 val &= ~INFINIPATH_HWE_PCIEBUSPARITYRADM;
595 }
577 dd->ipath_hwerrmask = val; 596 dd->ipath_hwerrmask = val;
578} 597}
579 598
@@ -583,8 +602,8 @@ static void ipath_pe_init_hwerrors(struct ipath_devdata *dd)
583 */ 602 */
584static int ipath_pe_bringup_serdes(struct ipath_devdata *dd) 603static int ipath_pe_bringup_serdes(struct ipath_devdata *dd)
585{ 604{
586 u64 val, tmp, config1; 605 u64 val, tmp, config1, prev_val;
587 int ret = 0, change = 0; 606 int ret = 0;
588 607
589 ipath_dbg("Trying to bringup serdes\n"); 608 ipath_dbg("Trying to bringup serdes\n");
590 609
@@ -641,6 +660,7 @@ static int ipath_pe_bringup_serdes(struct ipath_devdata *dd)
641 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); 660 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
642 661
643 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig); 662 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig);
663 prev_val = val;
644 if (((val >> INFINIPATH_XGXS_MDIOADDR_SHIFT) & 664 if (((val >> INFINIPATH_XGXS_MDIOADDR_SHIFT) &
645 INFINIPATH_XGXS_MDIOADDR_MASK) != 3) { 665 INFINIPATH_XGXS_MDIOADDR_MASK) != 3) {
646 val &= 666 val &=
@@ -648,11 +668,9 @@ static int ipath_pe_bringup_serdes(struct ipath_devdata *dd)
648 INFINIPATH_XGXS_MDIOADDR_SHIFT); 668 INFINIPATH_XGXS_MDIOADDR_SHIFT);
649 /* MDIO address 3 */ 669 /* MDIO address 3 */
650 val |= 3ULL << INFINIPATH_XGXS_MDIOADDR_SHIFT; 670 val |= 3ULL << INFINIPATH_XGXS_MDIOADDR_SHIFT;
651 change = 1;
652 } 671 }
653 if (val & INFINIPATH_XGXS_RESET) { 672 if (val & INFINIPATH_XGXS_RESET) {
654 val &= ~INFINIPATH_XGXS_RESET; 673 val &= ~INFINIPATH_XGXS_RESET;
655 change = 1;
656 } 674 }
657 if (((val >> INFINIPATH_XGXS_RX_POL_SHIFT) & 675 if (((val >> INFINIPATH_XGXS_RX_POL_SHIFT) &
658 INFINIPATH_XGXS_RX_POL_MASK) != dd->ipath_rx_pol_inv ) { 676 INFINIPATH_XGXS_RX_POL_MASK) != dd->ipath_rx_pol_inv ) {
@@ -661,9 +679,19 @@ static int ipath_pe_bringup_serdes(struct ipath_devdata *dd)
661 INFINIPATH_XGXS_RX_POL_SHIFT); 679 INFINIPATH_XGXS_RX_POL_SHIFT);
662 val |= dd->ipath_rx_pol_inv << 680 val |= dd->ipath_rx_pol_inv <<
663 INFINIPATH_XGXS_RX_POL_SHIFT; 681 INFINIPATH_XGXS_RX_POL_SHIFT;
664 change = 1;
665 } 682 }
666 if (change) 683 if (dd->ipath_minrev >= 2) {
684 /* Rev 2. can tolerate multiple writes to PBC, and
685 * allowing them can provide lower latency on some
686 * CPUs, but this feature is off by default, only
687 * turned on by setting D63 of XGXSconfig reg.
688 * May want to make this conditional more
689 * fine-grained in future. This is not exactly
690 * related to XGXS, but where the bit ended up.
691 */
692 val |= INFINIPATH_XGXS_SUPPRESS_ARMLAUNCH_ERR;
693 }
694 if (val != prev_val)
667 ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val); 695 ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val);
668 696
669 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig0); 697 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig0);
@@ -717,9 +745,25 @@ static void ipath_pe_quiet_serdes(struct ipath_devdata *dd)
717 ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val); 745 ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
718} 746}
719 747
720/* this is not yet needed on this chip, so just return 0. */
721static int ipath_pe_intconfig(struct ipath_devdata *dd) 748static int ipath_pe_intconfig(struct ipath_devdata *dd)
722{ 749{
750 u64 val;
751 u32 chiprev;
752
753 /*
754 * If the chip supports added error indication via GPIO pins,
755 * enable interrupts on those bits so the interrupt routine
756 * can count the events. Also set flag so interrupt routine
757 * can know they are expected.
758 */
759 chiprev = dd->ipath_revision >> INFINIPATH_R_CHIPREVMINOR_SHIFT;
760 if ((chiprev & INFINIPATH_R_CHIPREVMINOR_MASK) > 1) {
761 /* Rev2+ reports extra errors via internal GPIO pins */
762 dd->ipath_flags |= IPATH_GPIO_ERRINTRS;
763 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_gpio_mask);
764 val |= IPATH_GPIO_ERRINTR_MASK;
765 ipath_write_kreg( dd, dd->ipath_kregs->kr_gpio_mask, val);
766 }
723 return 0; 767 return 0;
724} 768}
725 769
@@ -853,21 +897,23 @@ static int ipath_setup_pe_config(struct ipath_devdata *dd,
853 return 0; 897 return 0;
854} 898}
855 899
856static void ipath_init_pe_variables(void) 900static void ipath_init_pe_variables(struct ipath_devdata *dd)
857{ 901{
858 /* 902 /*
859 * bits for selecting i2c direction and values, 903 * bits for selecting i2c direction and values,
860 * used for I2C serial flash 904 * used for I2C serial flash
861 */ 905 */
862 ipath_gpio_sda_num = _IPATH_GPIO_SDA_NUM; 906 dd->ipath_gpio_sda_num = _IPATH_GPIO_SDA_NUM;
863 ipath_gpio_scl_num = _IPATH_GPIO_SCL_NUM; 907 dd->ipath_gpio_scl_num = _IPATH_GPIO_SCL_NUM;
864 ipath_gpio_sda = IPATH_GPIO_SDA; 908 dd->ipath_gpio_sda = IPATH_GPIO_SDA;
865 ipath_gpio_scl = IPATH_GPIO_SCL; 909 dd->ipath_gpio_scl = IPATH_GPIO_SCL;
866 910
867 /* variables for sanity checking interrupt and errors */ 911 /* variables for sanity checking interrupt and errors */
868 infinipath_hwe_bitsextant = 912 dd->ipath_hwe_bitsextant =
869 (INFINIPATH_HWE_RXEMEMPARITYERR_MASK << 913 (INFINIPATH_HWE_RXEMEMPARITYERR_MASK <<
870 INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT) | 914 INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT) |
915 (INFINIPATH_HWE_TXEMEMPARITYERR_MASK <<
916 INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT) |
871 (INFINIPATH_HWE_PCIEMEMPARITYERR_MASK << 917 (INFINIPATH_HWE_PCIEMEMPARITYERR_MASK <<
872 INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT) | 918 INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT) |
873 INFINIPATH_HWE_PCIE1PLLFAILED | 919 INFINIPATH_HWE_PCIE1PLLFAILED |
@@ -883,13 +929,13 @@ static void ipath_init_pe_variables(void)
883 INFINIPATH_HWE_SERDESPLLFAILED | 929 INFINIPATH_HWE_SERDESPLLFAILED |
884 INFINIPATH_HWE_IBCBUSTOSPCPARITYERR | 930 INFINIPATH_HWE_IBCBUSTOSPCPARITYERR |
885 INFINIPATH_HWE_IBCBUSFRSPCPARITYERR; 931 INFINIPATH_HWE_IBCBUSFRSPCPARITYERR;
886 infinipath_i_bitsextant = 932 dd->ipath_i_bitsextant =
887 (INFINIPATH_I_RCVURG_MASK << INFINIPATH_I_RCVURG_SHIFT) | 933 (INFINIPATH_I_RCVURG_MASK << INFINIPATH_I_RCVURG_SHIFT) |
888 (INFINIPATH_I_RCVAVAIL_MASK << 934 (INFINIPATH_I_RCVAVAIL_MASK <<
889 INFINIPATH_I_RCVAVAIL_SHIFT) | 935 INFINIPATH_I_RCVAVAIL_SHIFT) |
890 INFINIPATH_I_ERROR | INFINIPATH_I_SPIOSENT | 936 INFINIPATH_I_ERROR | INFINIPATH_I_SPIOSENT |
891 INFINIPATH_I_SPIOBUFAVAIL | INFINIPATH_I_GPIO; 937 INFINIPATH_I_SPIOBUFAVAIL | INFINIPATH_I_GPIO;
892 infinipath_e_bitsextant = 938 dd->ipath_e_bitsextant =
893 INFINIPATH_E_RFORMATERR | INFINIPATH_E_RVCRC | 939 INFINIPATH_E_RFORMATERR | INFINIPATH_E_RVCRC |
894 INFINIPATH_E_RICRC | INFINIPATH_E_RMINPKTLEN | 940 INFINIPATH_E_RICRC | INFINIPATH_E_RMINPKTLEN |
895 INFINIPATH_E_RMAXPKTLEN | INFINIPATH_E_RLONGPKTLEN | 941 INFINIPATH_E_RMAXPKTLEN | INFINIPATH_E_RLONGPKTLEN |
@@ -907,8 +953,8 @@ static void ipath_init_pe_variables(void)
907 INFINIPATH_E_INVALIDADDR | INFINIPATH_E_RESET | 953 INFINIPATH_E_INVALIDADDR | INFINIPATH_E_RESET |
908 INFINIPATH_E_HARDWARE; 954 INFINIPATH_E_HARDWARE;
909 955
910 infinipath_i_rcvavail_mask = INFINIPATH_I_RCVAVAIL_MASK; 956 dd->ipath_i_rcvavail_mask = INFINIPATH_I_RCVAVAIL_MASK;
911 infinipath_i_rcvurg_mask = INFINIPATH_I_RCVURG_MASK; 957 dd->ipath_i_rcvurg_mask = INFINIPATH_I_RCVURG_MASK;
912} 958}
913 959
914/* setup the MSI stuff again after a reset. I'd like to just call 960/* setup the MSI stuff again after a reset. I'd like to just call
@@ -1082,6 +1128,45 @@ static void ipath_pe_put_tid(struct ipath_devdata *dd, u64 __iomem *tidptr,
1082 mmiowb(); 1128 mmiowb();
1083 spin_unlock_irqrestore(&dd->ipath_tid_lock, flags); 1129 spin_unlock_irqrestore(&dd->ipath_tid_lock, flags);
1084} 1130}
1131/**
1132 * ipath_pe_put_tid_2 - write a TID in chip, Revision 2 or higher
1133 * @dd: the infinipath device
1134 * @tidptr: pointer to the expected TID (in chip) to udpate
1135 * @tidtype: 0 for eager, 1 for expected
1136 * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing
1137 *
1138 * This exists as a separate routine to allow for selection of the
1139 * appropriate "flavor". The static calls in cleanup just use the
1140 * revision-agnostic form, as they are not performance critical.
1141 */
1142static void ipath_pe_put_tid_2(struct ipath_devdata *dd, u64 __iomem *tidptr,
1143 u32 type, unsigned long pa)
1144{
1145 u32 __iomem *tidp32 = (u32 __iomem *)tidptr;
1146
1147 if (pa != dd->ipath_tidinvalid) {
1148 if (pa & ((1U << 11) - 1)) {
1149 dev_info(&dd->pcidev->dev, "BUG: physaddr %lx "
1150 "not 2KB aligned!\n", pa);
1151 return;
1152 }
1153 pa >>= 11;
1154 /* paranoia check */
1155 if (pa & (7<<29))
1156 ipath_dev_err(dd,
1157 "BUG: Physical page address 0x%lx "
1158 "has bits set in 31-29\n", pa);
1159
1160 if (type == 0)
1161 pa |= dd->ipath_tidtemplate;
1162 else /* for now, always full 4KB page */
1163 pa |= 2 << 29;
1164 }
1165 if (dd->ipath_kregbase)
1166 writel(pa, tidp32);
1167 mmiowb();
1168}
1169
1085 1170
1086/** 1171/**
1087 * ipath_pe_clear_tid - clear all TID entries for a port, expected and eager 1172 * ipath_pe_clear_tid - clear all TID entries for a port, expected and eager
@@ -1203,7 +1288,7 @@ int __attribute__((weak)) ipath_unordered_wc(void)
1203 1288
1204/** 1289/**
1205 * ipath_init_pe_get_base_info - set chip-specific flags for user code 1290 * ipath_init_pe_get_base_info - set chip-specific flags for user code
1206 * @dd: the infinipath device 1291 * @pd: the infinipath port
1207 * @kbase: ipath_base_info pointer 1292 * @kbase: ipath_base_info pointer
1208 * 1293 *
1209 * We set the PCIE flag because the lower bandwidth on PCIe vs 1294 * We set the PCIE flag because the lower bandwidth on PCIe vs
@@ -1212,6 +1297,7 @@ int __attribute__((weak)) ipath_unordered_wc(void)
1212static int ipath_pe_get_base_info(struct ipath_portdata *pd, void *kbase) 1297static int ipath_pe_get_base_info(struct ipath_portdata *pd, void *kbase)
1213{ 1298{
1214 struct ipath_base_info *kinfo = kbase; 1299 struct ipath_base_info *kinfo = kbase;
1300 struct ipath_devdata *dd;
1215 1301
1216 if (ipath_unordered_wc()) { 1302 if (ipath_unordered_wc()) {
1217 kinfo->spi_runtime_flags |= IPATH_RUNTIME_FORCE_WC_ORDER; 1303 kinfo->spi_runtime_flags |= IPATH_RUNTIME_FORCE_WC_ORDER;
@@ -1220,8 +1306,20 @@ static int ipath_pe_get_base_info(struct ipath_portdata *pd, void *kbase)
1220 else 1306 else
1221 ipath_cdbg(PROC, "Not Intel processor, WC ordered\n"); 1307 ipath_cdbg(PROC, "Not Intel processor, WC ordered\n");
1222 1308
1223 kinfo->spi_runtime_flags |= IPATH_RUNTIME_PCIE; 1309 if (pd == NULL)
1310 goto done;
1224 1311
1312 dd = pd->port_dd;
1313
1314 if (dd != NULL && dd->ipath_minrev >= 2) {
1315 ipath_cdbg(PROC, "IBA6120 Rev2, allow multiple PBC write\n");
1316 kinfo->spi_runtime_flags |= IPATH_RUNTIME_PBC_REWRITE;
1317 ipath_cdbg(PROC, "IBA6120 Rev2, allow loose DMA alignment\n");
1318 kinfo->spi_runtime_flags |= IPATH_RUNTIME_LOOSE_DMA_ALIGN;
1319 }
1320
1321done:
1322 kinfo->spi_runtime_flags |= IPATH_RUNTIME_PCIE;
1225 return 0; 1323 return 0;
1226} 1324}
1227 1325
@@ -1244,7 +1342,10 @@ void ipath_init_iba6120_funcs(struct ipath_devdata *dd)
1244 dd->ipath_f_quiet_serdes = ipath_pe_quiet_serdes; 1342 dd->ipath_f_quiet_serdes = ipath_pe_quiet_serdes;
1245 dd->ipath_f_bringup_serdes = ipath_pe_bringup_serdes; 1343 dd->ipath_f_bringup_serdes = ipath_pe_bringup_serdes;
1246 dd->ipath_f_clear_tids = ipath_pe_clear_tids; 1344 dd->ipath_f_clear_tids = ipath_pe_clear_tids;
1247 dd->ipath_f_put_tid = ipath_pe_put_tid; 1345 if (dd->ipath_minrev >= 2)
1346 dd->ipath_f_put_tid = ipath_pe_put_tid_2;
1347 else
1348 dd->ipath_f_put_tid = ipath_pe_put_tid;
1248 dd->ipath_f_cleanup = ipath_setup_pe_cleanup; 1349 dd->ipath_f_cleanup = ipath_setup_pe_cleanup;
1249 dd->ipath_f_setextled = ipath_setup_pe_setextled; 1350 dd->ipath_f_setextled = ipath_setup_pe_setextled;
1250 dd->ipath_f_get_base_info = ipath_pe_get_base_info; 1351 dd->ipath_f_get_base_info = ipath_pe_get_base_info;
@@ -1259,6 +1360,6 @@ void ipath_init_iba6120_funcs(struct ipath_devdata *dd)
1259 dd->ipath_kregs = &ipath_pe_kregs; 1360 dd->ipath_kregs = &ipath_pe_kregs;
1260 dd->ipath_cregs = &ipath_pe_cregs; 1361 dd->ipath_cregs = &ipath_pe_cregs;
1261 1362
1262 ipath_init_pe_variables(); 1363 ipath_init_pe_variables(dd);
1263} 1364}
1264 1365
diff --git a/drivers/infiniband/hw/ipath/ipath_init_chip.c b/drivers/infiniband/hw/ipath/ipath_init_chip.c
index 44669dc2e22d..d819cca524cd 100644
--- a/drivers/infiniband/hw/ipath/ipath_init_chip.c
+++ b/drivers/infiniband/hw/ipath/ipath_init_chip.c
@@ -88,13 +88,13 @@ MODULE_PARM_DESC(kpiobufs, "Set number of PIO buffers for driver");
88static int create_port0_egr(struct ipath_devdata *dd) 88static int create_port0_egr(struct ipath_devdata *dd)
89{ 89{
90 unsigned e, egrcnt; 90 unsigned e, egrcnt;
91 struct sk_buff **skbs; 91 struct ipath_skbinfo *skbinfo;
92 int ret; 92 int ret;
93 93
94 egrcnt = dd->ipath_rcvegrcnt; 94 egrcnt = dd->ipath_rcvegrcnt;
95 95
96 skbs = vmalloc(sizeof(*dd->ipath_port0_skbs) * egrcnt); 96 skbinfo = vmalloc(sizeof(*dd->ipath_port0_skbinfo) * egrcnt);
97 if (skbs == NULL) { 97 if (skbinfo == NULL) {
98 ipath_dev_err(dd, "allocation error for eager TID " 98 ipath_dev_err(dd, "allocation error for eager TID "
99 "skb array\n"); 99 "skb array\n");
100 ret = -ENOMEM; 100 ret = -ENOMEM;
@@ -109,13 +109,13 @@ static int create_port0_egr(struct ipath_devdata *dd)
109 * 4 bytes so that the data buffer stays word aligned. 109 * 4 bytes so that the data buffer stays word aligned.
110 * See ipath_kreceive() for more details. 110 * See ipath_kreceive() for more details.
111 */ 111 */
112 skbs[e] = ipath_alloc_skb(dd, GFP_KERNEL); 112 skbinfo[e].skb = ipath_alloc_skb(dd, GFP_KERNEL);
113 if (!skbs[e]) { 113 if (!skbinfo[e].skb) {
114 ipath_dev_err(dd, "SKB allocation error for " 114 ipath_dev_err(dd, "SKB allocation error for "
115 "eager TID %u\n", e); 115 "eager TID %u\n", e);
116 while (e != 0) 116 while (e != 0)
117 dev_kfree_skb(skbs[--e]); 117 dev_kfree_skb(skbinfo[--e].skb);
118 vfree(skbs); 118 vfree(skbinfo);
119 ret = -ENOMEM; 119 ret = -ENOMEM;
120 goto bail; 120 goto bail;
121 } 121 }
@@ -124,14 +124,17 @@ static int create_port0_egr(struct ipath_devdata *dd)
124 * After loop above, so we can test non-NULL to see if ready 124 * After loop above, so we can test non-NULL to see if ready
125 * to use at receive, etc. 125 * to use at receive, etc.
126 */ 126 */
127 dd->ipath_port0_skbs = skbs; 127 dd->ipath_port0_skbinfo = skbinfo;
128 128
129 for (e = 0; e < egrcnt; e++) { 129 for (e = 0; e < egrcnt; e++) {
130 unsigned long phys = 130 dd->ipath_port0_skbinfo[e].phys =
131 virt_to_phys(dd->ipath_port0_skbs[e]->data); 131 ipath_map_single(dd->pcidev,
132 dd->ipath_port0_skbinfo[e].skb->data,
133 dd->ipath_ibmaxlen, PCI_DMA_FROMDEVICE);
132 dd->ipath_f_put_tid(dd, e + (u64 __iomem *) 134 dd->ipath_f_put_tid(dd, e + (u64 __iomem *)
133 ((char __iomem *) dd->ipath_kregbase + 135 ((char __iomem *) dd->ipath_kregbase +
134 dd->ipath_rcvegrbase), 0, phys); 136 dd->ipath_rcvegrbase), 0,
137 dd->ipath_port0_skbinfo[e].phys);
135 } 138 }
136 139
137 ret = 0; 140 ret = 0;
@@ -432,16 +435,33 @@ done:
432 */ 435 */
433static void init_shadow_tids(struct ipath_devdata *dd) 436static void init_shadow_tids(struct ipath_devdata *dd)
434{ 437{
435 dd->ipath_pageshadow = (struct page **) 438 struct page **pages;
436 vmalloc(dd->ipath_cfgports * dd->ipath_rcvtidcnt * 439 dma_addr_t *addrs;
440
441 pages = vmalloc(dd->ipath_cfgports * dd->ipath_rcvtidcnt *
437 sizeof(struct page *)); 442 sizeof(struct page *));
438 if (!dd->ipath_pageshadow) 443 if (!pages) {
439 ipath_dev_err(dd, "failed to allocate shadow page * " 444 ipath_dev_err(dd, "failed to allocate shadow page * "
440 "array, no expected sends!\n"); 445 "array, no expected sends!\n");
441 else 446 dd->ipath_pageshadow = NULL;
442 memset(dd->ipath_pageshadow, 0, 447 return;
443 dd->ipath_cfgports * dd->ipath_rcvtidcnt * 448 }
444 sizeof(struct page *)); 449
450 addrs = vmalloc(dd->ipath_cfgports * dd->ipath_rcvtidcnt *
451 sizeof(dma_addr_t));
452 if (!addrs) {
453 ipath_dev_err(dd, "failed to allocate shadow dma handle "
454 "array, no expected sends!\n");
455 vfree(dd->ipath_pageshadow);
456 dd->ipath_pageshadow = NULL;
457 return;
458 }
459
460 memset(pages, 0, dd->ipath_cfgports * dd->ipath_rcvtidcnt *
461 sizeof(struct page *));
462
463 dd->ipath_pageshadow = pages;
464 dd->ipath_physshadow = addrs;
445} 465}
446 466
447static void enable_chip(struct ipath_devdata *dd, 467static void enable_chip(struct ipath_devdata *dd,
diff --git a/drivers/infiniband/hw/ipath/ipath_intr.c b/drivers/infiniband/hw/ipath/ipath_intr.c
index 49bf7bb15b04..6bee53ce5f33 100644
--- a/drivers/infiniband/hw/ipath/ipath_intr.c
+++ b/drivers/infiniband/hw/ipath/ipath_intr.c
@@ -37,6 +37,50 @@
37#include "ipath_verbs.h" 37#include "ipath_verbs.h"
38#include "ipath_common.h" 38#include "ipath_common.h"
39 39
40/*
41 * Called when we might have an error that is specific to a particular
42 * PIO buffer, and may need to cancel that buffer, so it can be re-used.
43 */
44void ipath_disarm_senderrbufs(struct ipath_devdata *dd)
45{
46 u32 piobcnt;
47 unsigned long sbuf[4];
48 /*
49 * it's possible that sendbuffererror could have bits set; might
50 * have already done this as a result of hardware error handling
51 */
52 piobcnt = dd->ipath_piobcnt2k + dd->ipath_piobcnt4k;
53 /* read these before writing errorclear */
54 sbuf[0] = ipath_read_kreg64(
55 dd, dd->ipath_kregs->kr_sendbuffererror);
56 sbuf[1] = ipath_read_kreg64(
57 dd, dd->ipath_kregs->kr_sendbuffererror + 1);
58 if (piobcnt > 128) {
59 sbuf[2] = ipath_read_kreg64(
60 dd, dd->ipath_kregs->kr_sendbuffererror + 2);
61 sbuf[3] = ipath_read_kreg64(
62 dd, dd->ipath_kregs->kr_sendbuffererror + 3);
63 }
64
65 if (sbuf[0] || sbuf[1] || (piobcnt > 128 && (sbuf[2] || sbuf[3]))) {
66 int i;
67 if (ipath_debug & (__IPATH_PKTDBG|__IPATH_DBG)) {
68 __IPATH_DBG_WHICH(__IPATH_PKTDBG|__IPATH_DBG,
69 "SendbufErrs %lx %lx", sbuf[0],
70 sbuf[1]);
71 if (ipath_debug & __IPATH_PKTDBG && piobcnt > 128)
72 printk(" %lx %lx ", sbuf[2], sbuf[3]);
73 printk("\n");
74 }
75
76 for (i = 0; i < piobcnt; i++)
77 if (test_bit(i, sbuf))
78 ipath_disarm_piobufs(dd, i, 1);
79 dd->ipath_lastcancel = jiffies+3; /* no armlaunch for a bit */
80 }
81}
82
83
40/* These are all rcv-related errors which we want to count for stats */ 84/* These are all rcv-related errors which we want to count for stats */
41#define E_SUM_PKTERRS \ 85#define E_SUM_PKTERRS \
42 (INFINIPATH_E_RHDRLEN | INFINIPATH_E_RBADTID | \ 86 (INFINIPATH_E_RHDRLEN | INFINIPATH_E_RBADTID | \
@@ -68,53 +112,9 @@
68 112
69static u64 handle_e_sum_errs(struct ipath_devdata *dd, ipath_err_t errs) 113static u64 handle_e_sum_errs(struct ipath_devdata *dd, ipath_err_t errs)
70{ 114{
71 unsigned long sbuf[4];
72 u64 ignore_this_time = 0; 115 u64 ignore_this_time = 0;
73 u32 piobcnt;
74 116
75 /* if possible that sendbuffererror could be valid */ 117 ipath_disarm_senderrbufs(dd);
76 piobcnt = dd->ipath_piobcnt2k + dd->ipath_piobcnt4k;
77 /* read these before writing errorclear */
78 sbuf[0] = ipath_read_kreg64(
79 dd, dd->ipath_kregs->kr_sendbuffererror);
80 sbuf[1] = ipath_read_kreg64(
81 dd, dd->ipath_kregs->kr_sendbuffererror + 1);
82 if (piobcnt > 128) {
83 sbuf[2] = ipath_read_kreg64(
84 dd, dd->ipath_kregs->kr_sendbuffererror + 2);
85 sbuf[3] = ipath_read_kreg64(
86 dd, dd->ipath_kregs->kr_sendbuffererror + 3);
87 }
88
89 if (sbuf[0] || sbuf[1] || (piobcnt > 128 && (sbuf[2] || sbuf[3]))) {
90 int i;
91
92 ipath_cdbg(PKT, "SendbufErrs %lx %lx ", sbuf[0], sbuf[1]);
93 if (ipath_debug & __IPATH_PKTDBG && piobcnt > 128)
94 printk("%lx %lx ", sbuf[2], sbuf[3]);
95 for (i = 0; i < piobcnt; i++) {
96 if (test_bit(i, sbuf)) {
97 u32 __iomem *piobuf;
98 if (i < dd->ipath_piobcnt2k)
99 piobuf = (u32 __iomem *)
100 (dd->ipath_pio2kbase +
101 i * dd->ipath_palign);
102 else
103 piobuf = (u32 __iomem *)
104 (dd->ipath_pio4kbase +
105 (i - dd->ipath_piobcnt2k) *
106 dd->ipath_4kalign);
107
108 ipath_cdbg(PKT,
109 "PIObuf[%u] @%p pbc is %x; ",
110 i, piobuf, readl(piobuf));
111
112 ipath_disarm_piobufs(dd, i, 1);
113 }
114 }
115 if (ipath_debug & __IPATH_PKTDBG)
116 printk("\n");
117 }
118 if ((errs & E_SUM_LINK_PKTERRS) && 118 if ((errs & E_SUM_LINK_PKTERRS) &&
119 !(dd->ipath_flags & IPATH_LINKACTIVE)) { 119 !(dd->ipath_flags & IPATH_LINKACTIVE)) {
120 /* 120 /*
@@ -132,6 +132,82 @@ static u64 handle_e_sum_errs(struct ipath_devdata *dd, ipath_err_t errs)
132 return ignore_this_time; 132 return ignore_this_time;
133} 133}
134 134
135/* generic hw error messages... */
136#define INFINIPATH_HWE_TXEMEMPARITYERR_MSG(a) \
137 { \
138 .mask = ( INFINIPATH_HWE_TXEMEMPARITYERR_##a << \
139 INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT ), \
140 .msg = "TXE " #a " Memory Parity" \
141 }
142#define INFINIPATH_HWE_RXEMEMPARITYERR_MSG(a) \
143 { \
144 .mask = ( INFINIPATH_HWE_RXEMEMPARITYERR_##a << \
145 INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT ), \
146 .msg = "RXE " #a " Memory Parity" \
147 }
148
149static const struct ipath_hwerror_msgs ipath_generic_hwerror_msgs[] = {
150 INFINIPATH_HWE_MSG(IBCBUSFRSPCPARITYERR, "IPATH2IB Parity"),
151 INFINIPATH_HWE_MSG(IBCBUSTOSPCPARITYERR, "IB2IPATH Parity"),
152
153 INFINIPATH_HWE_TXEMEMPARITYERR_MSG(PIOBUF),
154 INFINIPATH_HWE_TXEMEMPARITYERR_MSG(PIOPBC),
155 INFINIPATH_HWE_TXEMEMPARITYERR_MSG(PIOLAUNCHFIFO),
156
157 INFINIPATH_HWE_RXEMEMPARITYERR_MSG(RCVBUF),
158 INFINIPATH_HWE_RXEMEMPARITYERR_MSG(LOOKUPQ),
159 INFINIPATH_HWE_RXEMEMPARITYERR_MSG(EAGERTID),
160 INFINIPATH_HWE_RXEMEMPARITYERR_MSG(EXPTID),
161 INFINIPATH_HWE_RXEMEMPARITYERR_MSG(FLAGBUF),
162 INFINIPATH_HWE_RXEMEMPARITYERR_MSG(DATAINFO),
163 INFINIPATH_HWE_RXEMEMPARITYERR_MSG(HDRINFO),
164};
165
166/**
167 * ipath_format_hwmsg - format a single hwerror message
168 * @msg message buffer
169 * @msgl length of message buffer
170 * @hwmsg message to add to message buffer
171 */
172static void ipath_format_hwmsg(char *msg, size_t msgl, const char *hwmsg)
173{
174 strlcat(msg, "[", msgl);
175 strlcat(msg, hwmsg, msgl);
176 strlcat(msg, "]", msgl);
177}
178
179/**
180 * ipath_format_hwerrors - format hardware error messages for display
181 * @hwerrs hardware errors bit vector
182 * @hwerrmsgs hardware error descriptions
183 * @nhwerrmsgs number of hwerrmsgs
184 * @msg message buffer
185 * @msgl message buffer length
186 */
187void ipath_format_hwerrors(u64 hwerrs,
188 const struct ipath_hwerror_msgs *hwerrmsgs,
189 size_t nhwerrmsgs,
190 char *msg, size_t msgl)
191{
192 int i;
193 const int glen =
194 sizeof(ipath_generic_hwerror_msgs) /
195 sizeof(ipath_generic_hwerror_msgs[0]);
196
197 for (i=0; i<glen; i++) {
198 if (hwerrs & ipath_generic_hwerror_msgs[i].mask) {
199 ipath_format_hwmsg(msg, msgl,
200 ipath_generic_hwerror_msgs[i].msg);
201 }
202 }
203
204 for (i=0; i<nhwerrmsgs; i++) {
205 if (hwerrs & hwerrmsgs[i].mask) {
206 ipath_format_hwmsg(msg, msgl, hwerrmsgs[i].msg);
207 }
208 }
209}
210
135/* return the strings for the most common link states */ 211/* return the strings for the most common link states */
136static char *ib_linkstate(u32 linkstate) 212static char *ib_linkstate(u32 linkstate)
137{ 213{
@@ -404,10 +480,10 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
404 dd->ipath_f_handle_hwerrors(dd, msg, sizeof msg); 480 dd->ipath_f_handle_hwerrors(dd, msg, sizeof msg);
405 } 481 }
406 482
407 if (!noprint && (errs & ~infinipath_e_bitsextant)) 483 if (!noprint && (errs & ~dd->ipath_e_bitsextant))
408 ipath_dev_err(dd, "error interrupt with unknown errors " 484 ipath_dev_err(dd, "error interrupt with unknown errors "
409 "%llx set\n", (unsigned long long) 485 "%llx set\n", (unsigned long long)
410 (errs & ~infinipath_e_bitsextant)); 486 (errs & ~dd->ipath_e_bitsextant));
411 487
412 if (errs & E_SUM_ERRS) 488 if (errs & E_SUM_ERRS)
413 ignore_this_time = handle_e_sum_errs(dd, errs); 489 ignore_this_time = handle_e_sum_errs(dd, errs);
@@ -478,6 +554,14 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
478 ~(INFINIPATH_E_HARDWARE | 554 ~(INFINIPATH_E_HARDWARE |
479 INFINIPATH_E_IBSTATUSCHANGED); 555 INFINIPATH_E_IBSTATUSCHANGED);
480 } 556 }
557
558 /* likely due to cancel, so suppress */
559 if ((errs & (INFINIPATH_E_SPKTLEN | INFINIPATH_E_SPIOARMLAUNCH)) &&
560 dd->ipath_lastcancel > jiffies) {
561 ipath_dbg("Suppressed armlaunch/spktlen after error send cancel\n");
562 errs &= ~(INFINIPATH_E_SPIOARMLAUNCH | INFINIPATH_E_SPKTLEN);
563 }
564
481 if (!errs) 565 if (!errs)
482 return 0; 566 return 0;
483 567
@@ -529,7 +613,7 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
529 * don't report same point multiple times, 613 * don't report same point multiple times,
530 * except kernel 614 * except kernel
531 */ 615 */
532 tl = (u32) * pd->port_rcvhdrtail_kvaddr; 616 tl = *(u64 *) pd->port_rcvhdrtail_kvaddr;
533 if (tl == dd->ipath_lastrcvhdrqtails[i]) 617 if (tl == dd->ipath_lastrcvhdrqtails[i])
534 continue; 618 continue;
535 hd = ipath_read_ureg32(dd, ur_rcvhdrhead, 619 hd = ipath_read_ureg32(dd, ur_rcvhdrhead,
@@ -729,9 +813,9 @@ static void handle_urcv(struct ipath_devdata *dd, u32 istat)
729 int rcvdint = 0; 813 int rcvdint = 0;
730 814
731 portr = ((istat >> INFINIPATH_I_RCVAVAIL_SHIFT) & 815 portr = ((istat >> INFINIPATH_I_RCVAVAIL_SHIFT) &
732 infinipath_i_rcvavail_mask) 816 dd->ipath_i_rcvavail_mask)
733 | ((istat >> INFINIPATH_I_RCVURG_SHIFT) & 817 | ((istat >> INFINIPATH_I_RCVURG_SHIFT) &
734 infinipath_i_rcvurg_mask); 818 dd->ipath_i_rcvurg_mask);
735 for (i = 1; i < dd->ipath_cfgports; i++) { 819 for (i = 1; i < dd->ipath_cfgports; i++) {
736 struct ipath_portdata *pd = dd->ipath_pd[i]; 820 struct ipath_portdata *pd = dd->ipath_pd[i];
737 if (portr & (1 << i) && pd && pd->port_cnt && 821 if (portr & (1 << i) && pd && pd->port_cnt &&
@@ -808,7 +892,7 @@ irqreturn_t ipath_intr(int irq, void *data, struct pt_regs *regs)
808 if (oldhead != curtail) { 892 if (oldhead != curtail) {
809 if (dd->ipath_flags & IPATH_GPIO_INTR) { 893 if (dd->ipath_flags & IPATH_GPIO_INTR) {
810 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_clear, 894 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_clear,
811 (u64) (1 << 2)); 895 (u64) (1 << IPATH_GPIO_PORT0_BIT));
812 istat = port0rbits | INFINIPATH_I_GPIO; 896 istat = port0rbits | INFINIPATH_I_GPIO;
813 } 897 }
814 else 898 else
@@ -838,10 +922,10 @@ irqreturn_t ipath_intr(int irq, void *data, struct pt_regs *regs)
838 if (unexpected) 922 if (unexpected)
839 unexpected = 0; 923 unexpected = 0;
840 924
841 if (unlikely(istat & ~infinipath_i_bitsextant)) 925 if (unlikely(istat & ~dd->ipath_i_bitsextant))
842 ipath_dev_err(dd, 926 ipath_dev_err(dd,
843 "interrupt with unknown interrupts %x set\n", 927 "interrupt with unknown interrupts %x set\n",
844 istat & (u32) ~ infinipath_i_bitsextant); 928 istat & (u32) ~ dd->ipath_i_bitsextant);
845 else 929 else
846 ipath_cdbg(VERBOSE, "intr stat=0x%x\n", istat); 930 ipath_cdbg(VERBOSE, "intr stat=0x%x\n", istat);
847 931
@@ -867,26 +951,80 @@ irqreturn_t ipath_intr(int irq, void *data, struct pt_regs *regs)
867 951
868 if (istat & INFINIPATH_I_GPIO) { 952 if (istat & INFINIPATH_I_GPIO) {
869 /* 953 /*
870 * Packets are available in the port 0 rcv queue. 954 * GPIO interrupts fall in two broad classes:
871 * Eventually this needs to be generalized to check 955 * GPIO_2 indicates (on some HT4xx boards) that a packet
872 * IPATH_GPIO_INTR, and the specific GPIO bit, if 956 * has arrived for Port 0. Checking for this
873 * GPIO interrupts are used for anything else. 957 * is controlled by flag IPATH_GPIO_INTR.
958 * GPIO_3..5 on IBA6120 Rev2 chips indicate errors
959 * that we need to count. Checking for this
960 * is controlled by flag IPATH_GPIO_ERRINTRS.
874 */ 961 */
875 if (unlikely(!(dd->ipath_flags & IPATH_GPIO_INTR))) { 962 u32 gpiostatus;
876 u32 gpiostatus; 963 u32 to_clear = 0;
877 gpiostatus = ipath_read_kreg32( 964
878 dd, dd->ipath_kregs->kr_gpio_status); 965 gpiostatus = ipath_read_kreg32(
879 ipath_dbg("Unexpected GPIO interrupt bits %x\n", 966 dd, dd->ipath_kregs->kr_gpio_status);
880 gpiostatus); 967 /* First the error-counter case.
881 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_clear, 968 */
882 gpiostatus); 969 if ((gpiostatus & IPATH_GPIO_ERRINTR_MASK) &&
970 (dd->ipath_flags & IPATH_GPIO_ERRINTRS)) {
971 /* want to clear the bits we see asserted. */
972 to_clear |= (gpiostatus & IPATH_GPIO_ERRINTR_MASK);
973
974 /*
975 * Count appropriately, clear bits out of our copy,
976 * as they have been "handled".
977 */
978 if (gpiostatus & (1 << IPATH_GPIO_RXUVL_BIT)) {
979 ipath_dbg("FlowCtl on UnsupVL\n");
980 dd->ipath_rxfc_unsupvl_errs++;
981 }
982 if (gpiostatus & (1 << IPATH_GPIO_OVRUN_BIT)) {
983 ipath_dbg("Overrun Threshold exceeded\n");
984 dd->ipath_overrun_thresh_errs++;
985 }
986 if (gpiostatus & (1 << IPATH_GPIO_LLI_BIT)) {
987 ipath_dbg("Local Link Integrity error\n");
988 dd->ipath_lli_errs++;
989 }
990 gpiostatus &= ~IPATH_GPIO_ERRINTR_MASK;
883 } 991 }
884 else { 992 /* Now the Port0 Receive case */
885 /* Clear GPIO status bit 2 */ 993 if ((gpiostatus & (1 << IPATH_GPIO_PORT0_BIT)) &&
886 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_clear, 994 (dd->ipath_flags & IPATH_GPIO_INTR)) {
887 (u64) (1 << 2)); 995 /*
996 * GPIO status bit 2 is set, and we expected it.
997 * clear it and indicate in p0bits.
998 * This probably only happens if a Port0 pkt
999 * arrives at _just_ the wrong time, and we
1000 * handle that by seting chk0rcv;
1001 */
1002 to_clear |= (1 << IPATH_GPIO_PORT0_BIT);
1003 gpiostatus &= ~(1 << IPATH_GPIO_PORT0_BIT);
888 chk0rcv = 1; 1004 chk0rcv = 1;
889 } 1005 }
1006 if (unlikely(gpiostatus)) {
1007 /*
1008 * Some unexpected bits remain. If they could have
1009 * caused the interrupt, complain and clear.
1010 * MEA: this is almost certainly non-ideal.
1011 * we should look into auto-disable of unexpected
1012 * GPIO interrupts, possibly on a "three strikes"
1013 * basis.
1014 */
1015 u32 mask;
1016 mask = ipath_read_kreg32(
1017 dd, dd->ipath_kregs->kr_gpio_mask);
1018 if (mask & gpiostatus) {
1019 ipath_dbg("Unexpected GPIO IRQ bits %x\n",
1020 gpiostatus & mask);
1021 to_clear |= (gpiostatus & mask);
1022 }
1023 }
1024 if (to_clear) {
1025 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_clear,
1026 (u64) to_clear);
1027 }
890 } 1028 }
891 chk0rcv |= istat & port0rbits; 1029 chk0rcv |= istat & port0rbits;
892 1030
@@ -911,9 +1049,9 @@ irqreturn_t ipath_intr(int irq, void *data, struct pt_regs *regs)
911 istat &= ~port0rbits; 1049 istat &= ~port0rbits;
912 } 1050 }
913 1051
914 if (istat & ((infinipath_i_rcvavail_mask << 1052 if (istat & ((dd->ipath_i_rcvavail_mask <<
915 INFINIPATH_I_RCVAVAIL_SHIFT) 1053 INFINIPATH_I_RCVAVAIL_SHIFT)
916 | (infinipath_i_rcvurg_mask << 1054 | (dd->ipath_i_rcvurg_mask <<
917 INFINIPATH_I_RCVURG_SHIFT))) 1055 INFINIPATH_I_RCVURG_SHIFT)))
918 handle_urcv(dd, istat); 1056 handle_urcv(dd, istat);
919 1057
diff --git a/drivers/infiniband/hw/ipath/ipath_kernel.h b/drivers/infiniband/hw/ipath/ipath_kernel.h
index a8a56276ff1d..d7540b71b451 100644
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h
@@ -39,6 +39,8 @@
39 */ 39 */
40 40
41#include <linux/interrupt.h> 41#include <linux/interrupt.h>
42#include <linux/pci.h>
43#include <linux/dma-mapping.h>
42#include <asm/io.h> 44#include <asm/io.h>
43 45
44#include "ipath_common.h" 46#include "ipath_common.h"
@@ -62,7 +64,7 @@ struct ipath_portdata {
62 /* rcvhdrq base, needs mmap before useful */ 64 /* rcvhdrq base, needs mmap before useful */
63 void *port_rcvhdrq; 65 void *port_rcvhdrq;
64 /* kernel virtual address where hdrqtail is updated */ 66 /* kernel virtual address where hdrqtail is updated */
65 volatile __le64 *port_rcvhdrtail_kvaddr; 67 void *port_rcvhdrtail_kvaddr;
66 /* 68 /*
67 * temp buffer for expected send setup, allocated at open, instead 69 * temp buffer for expected send setup, allocated at open, instead
68 * of each setup call 70 * of each setup call
@@ -79,8 +81,8 @@ struct ipath_portdata {
79 dma_addr_t port_rcvhdrq_phys; 81 dma_addr_t port_rcvhdrq_phys;
80 dma_addr_t port_rcvhdrqtailaddr_phys; 82 dma_addr_t port_rcvhdrqtailaddr_phys;
81 /* 83 /*
82 * number of opens on this instance (0 or 1; ignoring forks, dup, 84 * number of opens (including slave subports) on this instance
83 * etc. for now) 85 * (ignoring forks, dup, etc. for now)
84 */ 86 */
85 int port_cnt; 87 int port_cnt;
86 /* 88 /*
@@ -89,6 +91,10 @@ struct ipath_portdata {
89 */ 91 */
90 /* instead of calculating it */ 92 /* instead of calculating it */
91 unsigned port_port; 93 unsigned port_port;
94 /* non-zero if port is being shared. */
95 u16 port_subport_cnt;
96 /* non-zero if port is being shared. */
97 u16 port_subport_id;
92 /* chip offset of PIO buffers for this port */ 98 /* chip offset of PIO buffers for this port */
93 u32 port_piobufs; 99 u32 port_piobufs;
94 /* how many alloc_pages() chunks in port_rcvegrbuf_pages */ 100 /* how many alloc_pages() chunks in port_rcvegrbuf_pages */
@@ -121,6 +127,16 @@ struct ipath_portdata {
121 u16 port_pkeys[4]; 127 u16 port_pkeys[4];
122 /* so file ops can get at unit */ 128 /* so file ops can get at unit */
123 struct ipath_devdata *port_dd; 129 struct ipath_devdata *port_dd;
130 /* A page of memory for rcvhdrhead, rcvegrhead, rcvegrtail * N */
131 void *subport_uregbase;
132 /* An array of pages for the eager receive buffers * N */
133 void *subport_rcvegrbuf;
134 /* An array of pages for the eager header queue entries * N */
135 void *subport_rcvhdr_base;
136 /* The version of the library which opened this port */
137 u32 userversion;
138 /* Bitmask of active slaves */
139 u32 active_slaves;
124}; 140};
125 141
126struct sk_buff; 142struct sk_buff;
@@ -132,6 +148,11 @@ struct _ipath_layer {
132 void *l_arg; 148 void *l_arg;
133}; 149};
134 150
151struct ipath_skbinfo {
152 struct sk_buff *skb;
153 dma_addr_t phys;
154};
155
135struct ipath_devdata { 156struct ipath_devdata {
136 struct list_head ipath_list; 157 struct list_head ipath_list;
137 158
@@ -154,7 +175,7 @@ struct ipath_devdata {
154 /* ipath_cfgports pointers */ 175 /* ipath_cfgports pointers */
155 struct ipath_portdata **ipath_pd; 176 struct ipath_portdata **ipath_pd;
156 /* sk_buffs used by port 0 eager receive queue */ 177 /* sk_buffs used by port 0 eager receive queue */
157 struct sk_buff **ipath_port0_skbs; 178 struct ipath_skbinfo *ipath_port0_skbinfo;
158 /* kvirt address of 1st 2k pio buffer */ 179 /* kvirt address of 1st 2k pio buffer */
159 void __iomem *ipath_pio2kbase; 180 void __iomem *ipath_pio2kbase;
160 /* kvirt address of 1st 4k pio buffer */ 181 /* kvirt address of 1st 4k pio buffer */
@@ -315,12 +336,16 @@ struct ipath_devdata {
315 u8 ipath_ht_slave_off; 336 u8 ipath_ht_slave_off;
316 /* for write combining settings */ 337 /* for write combining settings */
317 unsigned long ipath_wc_cookie; 338 unsigned long ipath_wc_cookie;
339 unsigned long ipath_wc_base;
340 unsigned long ipath_wc_len;
318 /* ref count for each pkey */ 341 /* ref count for each pkey */
319 atomic_t ipath_pkeyrefs[4]; 342 atomic_t ipath_pkeyrefs[4];
320 /* shadow copy of all exptids physaddr; used only by funcsim */ 343 /* shadow copy of all exptids physaddr; used only by funcsim */
321 u64 *ipath_tidsimshadow; 344 u64 *ipath_tidsimshadow;
322 /* shadow copy of struct page *'s for exp tid pages */ 345 /* shadow copy of struct page *'s for exp tid pages */
323 struct page **ipath_pageshadow; 346 struct page **ipath_pageshadow;
347 /* shadow copy of dma handles for exp tid pages */
348 dma_addr_t *ipath_physshadow;
324 /* lock to workaround chip bug 9437 */ 349 /* lock to workaround chip bug 9437 */
325 spinlock_t ipath_tid_lock; 350 spinlock_t ipath_tid_lock;
326 351
@@ -402,6 +427,9 @@ struct ipath_devdata {
402 unsigned long ipath_rcvctrl; 427 unsigned long ipath_rcvctrl;
403 /* shadow kr_sendctrl */ 428 /* shadow kr_sendctrl */
404 unsigned long ipath_sendctrl; 429 unsigned long ipath_sendctrl;
430 /* ports waiting for PIOavail intr */
431 unsigned long ipath_portpiowait;
432 unsigned long ipath_lastcancel; /* to not count armlaunch after cancel */
405 433
406 /* value we put in kr_rcvhdrcnt */ 434 /* value we put in kr_rcvhdrcnt */
407 u32 ipath_rcvhdrcnt; 435 u32 ipath_rcvhdrcnt;
@@ -465,8 +493,6 @@ struct ipath_devdata {
465 u32 ipath_htwidth; 493 u32 ipath_htwidth;
466 /* HT speed (200,400,800,1000) from HT config */ 494 /* HT speed (200,400,800,1000) from HT config */
467 u32 ipath_htspeed; 495 u32 ipath_htspeed;
468 /* ports waiting for PIOavail intr */
469 unsigned long ipath_portpiowait;
470 /* 496 /*
471 * number of sequential ibcstatus change for polling active/quiet 497 * number of sequential ibcstatus change for polling active/quiet
472 * (i.e., link not coming up). 498 * (i.e., link not coming up).
@@ -510,8 +536,47 @@ struct ipath_devdata {
510 u32 ipath_lli_counter; 536 u32 ipath_lli_counter;
511 /* local link integrity errors */ 537 /* local link integrity errors */
512 u32 ipath_lli_errors; 538 u32 ipath_lli_errors;
539 /*
540 * Above counts only cases where _successive_ LocalLinkIntegrity
541 * errors were seen in the receive headers of kern-packets.
542 * Below are the three (monotonically increasing) counters
543 * maintained via GPIO interrupts on iba6120-rev2.
544 */
545 u32 ipath_rxfc_unsupvl_errs;
546 u32 ipath_overrun_thresh_errs;
547 u32 ipath_lli_errs;
548
549 /*
550 * Not all devices managed by a driver instance are the same
551 * type, so these fields must be per-device.
552 */
553 u64 ipath_i_bitsextant;
554 ipath_err_t ipath_e_bitsextant;
555 ipath_err_t ipath_hwe_bitsextant;
556
557 /*
558 * Below should be computable from number of ports,
559 * since they are never modified.
560 */
561 u32 ipath_i_rcvavail_mask;
562 u32 ipath_i_rcvurg_mask;
563
564 /*
565 * Register bits for selecting i2c direction and values, used for
566 * I2C serial flash.
567 */
568 u16 ipath_gpio_sda_num;
569 u16 ipath_gpio_scl_num;
570 u64 ipath_gpio_sda;
571 u64 ipath_gpio_scl;
513}; 572};
514 573
574/* Private data for file operations */
575struct ipath_filedata {
576 struct ipath_portdata *pd;
577 unsigned subport;
578 unsigned tidcursor;
579};
515extern struct list_head ipath_dev_list; 580extern struct list_head ipath_dev_list;
516extern spinlock_t ipath_devs_lock; 581extern spinlock_t ipath_devs_lock;
517extern struct ipath_devdata *ipath_lookup(int unit); 582extern struct ipath_devdata *ipath_lookup(int unit);
@@ -521,6 +586,7 @@ int ipath_enable_wc(struct ipath_devdata *dd);
521void ipath_disable_wc(struct ipath_devdata *dd); 586void ipath_disable_wc(struct ipath_devdata *dd);
522int ipath_count_units(int *npresentp, int *nupp, u32 *maxportsp); 587int ipath_count_units(int *npresentp, int *nupp, u32 *maxportsp);
523void ipath_shutdown_device(struct ipath_devdata *); 588void ipath_shutdown_device(struct ipath_devdata *);
589void ipath_disarm_senderrbufs(struct ipath_devdata *);
524 590
525struct file_operations; 591struct file_operations;
526int ipath_cdev_init(int minor, char *name, struct file_operations *fops, 592int ipath_cdev_init(int minor, char *name, struct file_operations *fops,
@@ -572,7 +638,11 @@ int ipath_set_lid(struct ipath_devdata *, u32, u8);
572int ipath_set_rx_pol_inv(struct ipath_devdata *dd, u8 new_pol_inv); 638int ipath_set_rx_pol_inv(struct ipath_devdata *dd, u8 new_pol_inv);
573 639
574/* for use in system calls, where we want to know device type, etc. */ 640/* for use in system calls, where we want to know device type, etc. */
575#define port_fp(fp) ((struct ipath_portdata *) (fp)->private_data) 641#define port_fp(fp) ((struct ipath_filedata *)(fp)->private_data)->pd
642#define subport_fp(fp) \
643 ((struct ipath_filedata *)(fp)->private_data)->subport
644#define tidcursor_fp(fp) \
645 ((struct ipath_filedata *)(fp)->private_data)->tidcursor
576 646
577/* 647/*
578 * values for ipath_flags 648 * values for ipath_flags
@@ -612,6 +682,15 @@ int ipath_set_rx_pol_inv(struct ipath_devdata *dd, u8 new_pol_inv);
612 /* can miss port0 rx interrupts */ 682 /* can miss port0 rx interrupts */
613#define IPATH_POLL_RX_INTR 0x40000 683#define IPATH_POLL_RX_INTR 0x40000
614#define IPATH_DISABLED 0x80000 /* administratively disabled */ 684#define IPATH_DISABLED 0x80000 /* administratively disabled */
685 /* Use GPIO interrupts for new counters */
686#define IPATH_GPIO_ERRINTRS 0x100000
687
688/* Bits in GPIO for the added interrupts */
689#define IPATH_GPIO_PORT0_BIT 2
690#define IPATH_GPIO_RXUVL_BIT 3
691#define IPATH_GPIO_OVRUN_BIT 4
692#define IPATH_GPIO_LLI_BIT 5
693#define IPATH_GPIO_ERRINTR_MASK 0x38
615 694
616/* portdata flag bit offsets */ 695/* portdata flag bit offsets */
617 /* waiting for a packet to arrive */ 696 /* waiting for a packet to arrive */
@@ -799,6 +878,13 @@ int ipathfs_add_device(struct ipath_devdata *);
799int ipathfs_remove_device(struct ipath_devdata *); 878int ipathfs_remove_device(struct ipath_devdata *);
800 879
801/* 880/*
881 * dma_addr wrappers - all 0's invalid for hw
882 */
883dma_addr_t ipath_map_page(struct pci_dev *, struct page *, unsigned long,
884 size_t, int);
885dma_addr_t ipath_map_single(struct pci_dev *, void *, size_t, int);
886
887/*
802 * Flush write combining store buffers (if present) and perform a write 888 * Flush write combining store buffers (if present) and perform a write
803 * barrier. 889 * barrier.
804 */ 890 */
@@ -855,4 +941,20 @@ extern struct mutex ipath_mutex;
855 941
856#endif /* _IPATH_DEBUGGING */ 942#endif /* _IPATH_DEBUGGING */
857 943
944/*
945 * this is used for formatting hw error messages...
946 */
947struct ipath_hwerror_msgs {
948 u64 mask;
949 const char *msg;
950};
951
952#define INFINIPATH_HWE_MSG(a, b) { .mask = INFINIPATH_HWE_##a, .msg = b }
953
954/* in ipath_intr.c... */
955void ipath_format_hwerrors(u64 hwerrs,
956 const struct ipath_hwerror_msgs *hwerrmsgs,
957 size_t nhwerrmsgs,
958 char *msg, size_t lmsg);
959
858#endif /* _IPATH_KERNEL_H */ 960#endif /* _IPATH_KERNEL_H */
diff --git a/drivers/infiniband/hw/ipath/ipath_keys.c b/drivers/infiniband/hw/ipath/ipath_keys.c
index ba1b93226caa..9a6cbd05adcd 100644
--- a/drivers/infiniband/hw/ipath/ipath_keys.c
+++ b/drivers/infiniband/hw/ipath/ipath_keys.c
@@ -118,9 +118,10 @@ void ipath_free_lkey(struct ipath_lkey_table *rkt, u32 lkey)
118 * Check the IB SGE for validity and initialize our internal version 118 * Check the IB SGE for validity and initialize our internal version
119 * of it. 119 * of it.
120 */ 120 */
121int ipath_lkey_ok(struct ipath_lkey_table *rkt, struct ipath_sge *isge, 121int ipath_lkey_ok(struct ipath_qp *qp, struct ipath_sge *isge,
122 struct ib_sge *sge, int acc) 122 struct ib_sge *sge, int acc)
123{ 123{
124 struct ipath_lkey_table *rkt = &to_idev(qp->ibqp.device)->lk_table;
124 struct ipath_mregion *mr; 125 struct ipath_mregion *mr;
125 unsigned n, m; 126 unsigned n, m;
126 size_t off; 127 size_t off;
@@ -140,7 +141,8 @@ int ipath_lkey_ok(struct ipath_lkey_table *rkt, struct ipath_sge *isge,
140 goto bail; 141 goto bail;
141 } 142 }
142 mr = rkt->table[(sge->lkey >> (32 - ib_ipath_lkey_table_size))]; 143 mr = rkt->table[(sge->lkey >> (32 - ib_ipath_lkey_table_size))];
143 if (unlikely(mr == NULL || mr->lkey != sge->lkey)) { 144 if (unlikely(mr == NULL || mr->lkey != sge->lkey ||
145 qp->ibqp.pd != mr->pd)) {
144 ret = 0; 146 ret = 0;
145 goto bail; 147 goto bail;
146 } 148 }
@@ -188,9 +190,10 @@ bail:
188 * 190 *
189 * Return 1 if successful, otherwise 0. 191 * Return 1 if successful, otherwise 0.
190 */ 192 */
191int ipath_rkey_ok(struct ipath_ibdev *dev, struct ipath_sge_state *ss, 193int ipath_rkey_ok(struct ipath_qp *qp, struct ipath_sge_state *ss,
192 u32 len, u64 vaddr, u32 rkey, int acc) 194 u32 len, u64 vaddr, u32 rkey, int acc)
193{ 195{
196 struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
194 struct ipath_lkey_table *rkt = &dev->lk_table; 197 struct ipath_lkey_table *rkt = &dev->lk_table;
195 struct ipath_sge *sge = &ss->sge; 198 struct ipath_sge *sge = &ss->sge;
196 struct ipath_mregion *mr; 199 struct ipath_mregion *mr;
@@ -214,7 +217,8 @@ int ipath_rkey_ok(struct ipath_ibdev *dev, struct ipath_sge_state *ss,
214 } 217 }
215 218
216 mr = rkt->table[(rkey >> (32 - ib_ipath_lkey_table_size))]; 219 mr = rkt->table[(rkey >> (32 - ib_ipath_lkey_table_size))];
217 if (unlikely(mr == NULL || mr->lkey != rkey)) { 220 if (unlikely(mr == NULL || mr->lkey != rkey ||
221 qp->ibqp.pd != mr->pd)) {
218 ret = 0; 222 ret = 0;
219 goto bail; 223 goto bail;
220 } 224 }
diff --git a/drivers/infiniband/hw/ipath/ipath_mad.c b/drivers/infiniband/hw/ipath/ipath_mad.c
index 72d1db89db8f..25908b02fbe5 100644
--- a/drivers/infiniband/hw/ipath/ipath_mad.c
+++ b/drivers/infiniband/hw/ipath/ipath_mad.c
@@ -87,7 +87,8 @@ static int recv_subn_get_nodeinfo(struct ib_smp *smp,
87 struct ipath_devdata *dd = to_idev(ibdev)->dd; 87 struct ipath_devdata *dd = to_idev(ibdev)->dd;
88 u32 vendor, majrev, minrev; 88 u32 vendor, majrev, minrev;
89 89
90 if (smp->attr_mod) 90 /* GUID 0 is illegal */
91 if (smp->attr_mod || (dd->ipath_guid == 0))
91 smp->status |= IB_SMP_INVALID_FIELD; 92 smp->status |= IB_SMP_INVALID_FIELD;
92 93
93 nip->base_version = 1; 94 nip->base_version = 1;
@@ -131,10 +132,15 @@ static int recv_subn_get_guidinfo(struct ib_smp *smp,
131 * We only support one GUID for now. If this changes, the 132 * We only support one GUID for now. If this changes, the
132 * portinfo.guid_cap field needs to be updated too. 133 * portinfo.guid_cap field needs to be updated too.
133 */ 134 */
134 if (startgx == 0) 135 if (startgx == 0) {
135 /* The first is a copy of the read-only HW GUID. */ 136 __be64 g = to_idev(ibdev)->dd->ipath_guid;
136 *p = to_idev(ibdev)->dd->ipath_guid; 137 if (g == 0)
137 else 138 /* GUID 0 is illegal */
139 smp->status |= IB_SMP_INVALID_FIELD;
140 else
141 /* The first is a copy of the read-only HW GUID. */
142 *p = g;
143 } else
138 smp->status |= IB_SMP_INVALID_FIELD; 144 smp->status |= IB_SMP_INVALID_FIELD;
139 145
140 return reply(smp); 146 return reply(smp);
diff --git a/drivers/infiniband/hw/ipath/ipath_mr.c b/drivers/infiniband/hw/ipath/ipath_mr.c
index b36f6fb3e37a..a0673c1eef71 100644
--- a/drivers/infiniband/hw/ipath/ipath_mr.c
+++ b/drivers/infiniband/hw/ipath/ipath_mr.c
@@ -138,6 +138,7 @@ struct ib_mr *ipath_reg_phys_mr(struct ib_pd *pd,
138 goto bail; 138 goto bail;
139 } 139 }
140 140
141 mr->mr.pd = pd;
141 mr->mr.user_base = *iova_start; 142 mr->mr.user_base = *iova_start;
142 mr->mr.iova = *iova_start; 143 mr->mr.iova = *iova_start;
143 mr->mr.length = 0; 144 mr->mr.length = 0;
@@ -197,6 +198,7 @@ struct ib_mr *ipath_reg_user_mr(struct ib_pd *pd, struct ib_umem *region,
197 goto bail; 198 goto bail;
198 } 199 }
199 200
201 mr->mr.pd = pd;
200 mr->mr.user_base = region->user_base; 202 mr->mr.user_base = region->user_base;
201 mr->mr.iova = region->virt_base; 203 mr->mr.iova = region->virt_base;
202 mr->mr.length = region->length; 204 mr->mr.length = region->length;
@@ -289,6 +291,7 @@ struct ib_fmr *ipath_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
289 * Resources are allocated but no valid mapping (RKEY can't be 291 * Resources are allocated but no valid mapping (RKEY can't be
290 * used). 292 * used).
291 */ 293 */
294 fmr->mr.pd = pd;
292 fmr->mr.user_base = 0; 295 fmr->mr.user_base = 0;
293 fmr->mr.iova = 0; 296 fmr->mr.iova = 0;
294 fmr->mr.length = 0; 297 fmr->mr.length = 0;
diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c
index 224b0f40767f..46c1c89bf6ae 100644
--- a/drivers/infiniband/hw/ipath/ipath_qp.c
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c
@@ -335,6 +335,7 @@ static void ipath_reset_qp(struct ipath_qp *qp)
335 qp->s_ack_state = IB_OPCODE_RC_ACKNOWLEDGE; 335 qp->s_ack_state = IB_OPCODE_RC_ACKNOWLEDGE;
336 qp->r_ack_state = IB_OPCODE_RC_ACKNOWLEDGE; 336 qp->r_ack_state = IB_OPCODE_RC_ACKNOWLEDGE;
337 qp->r_nak_state = 0; 337 qp->r_nak_state = 0;
338 qp->r_wrid_valid = 0;
338 qp->s_rnr_timeout = 0; 339 qp->s_rnr_timeout = 0;
339 qp->s_head = 0; 340 qp->s_head = 0;
340 qp->s_tail = 0; 341 qp->s_tail = 0;
@@ -342,6 +343,7 @@ static void ipath_reset_qp(struct ipath_qp *qp)
342 qp->s_last = 0; 343 qp->s_last = 0;
343 qp->s_ssn = 1; 344 qp->s_ssn = 1;
344 qp->s_lsn = 0; 345 qp->s_lsn = 0;
346 qp->s_wait_credit = 0;
345 if (qp->r_rq.wq) { 347 if (qp->r_rq.wq) {
346 qp->r_rq.wq->head = 0; 348 qp->r_rq.wq->head = 0;
347 qp->r_rq.wq->tail = 0; 349 qp->r_rq.wq->tail = 0;
@@ -352,12 +354,13 @@ static void ipath_reset_qp(struct ipath_qp *qp)
352/** 354/**
353 * ipath_error_qp - put a QP into an error state 355 * ipath_error_qp - put a QP into an error state
354 * @qp: the QP to put into an error state 356 * @qp: the QP to put into an error state
357 * @err: the receive completion error to signal if a RWQE is active
355 * 358 *
356 * Flushes both send and receive work queues. 359 * Flushes both send and receive work queues.
357 * QP s_lock should be held and interrupts disabled. 360 * QP s_lock should be held and interrupts disabled.
358 */ 361 */
359 362
360void ipath_error_qp(struct ipath_qp *qp) 363void ipath_error_qp(struct ipath_qp *qp, enum ib_wc_status err)
361{ 364{
362 struct ipath_ibdev *dev = to_idev(qp->ibqp.device); 365 struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
363 struct ib_wc wc; 366 struct ib_wc wc;
@@ -373,7 +376,6 @@ void ipath_error_qp(struct ipath_qp *qp)
373 list_del_init(&qp->piowait); 376 list_del_init(&qp->piowait);
374 spin_unlock(&dev->pending_lock); 377 spin_unlock(&dev->pending_lock);
375 378
376 wc.status = IB_WC_WR_FLUSH_ERR;
377 wc.vendor_err = 0; 379 wc.vendor_err = 0;
378 wc.byte_len = 0; 380 wc.byte_len = 0;
379 wc.imm_data = 0; 381 wc.imm_data = 0;
@@ -385,6 +387,12 @@ void ipath_error_qp(struct ipath_qp *qp)
385 wc.sl = 0; 387 wc.sl = 0;
386 wc.dlid_path_bits = 0; 388 wc.dlid_path_bits = 0;
387 wc.port_num = 0; 389 wc.port_num = 0;
390 if (qp->r_wrid_valid) {
391 qp->r_wrid_valid = 0;
392 wc.status = err;
393 ipath_cq_enter(to_icq(qp->ibqp.send_cq), &wc, 1);
394 }
395 wc.status = IB_WC_WR_FLUSH_ERR;
388 396
389 while (qp->s_last != qp->s_head) { 397 while (qp->s_last != qp->s_head) {
390 struct ipath_swqe *wqe = get_swqe_ptr(qp, qp->s_last); 398 struct ipath_swqe *wqe = get_swqe_ptr(qp, qp->s_last);
@@ -501,7 +509,7 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
501 break; 509 break;
502 510
503 case IB_QPS_ERR: 511 case IB_QPS_ERR:
504 ipath_error_qp(qp); 512 ipath_error_qp(qp, IB_WC_GENERAL_ERR);
505 break; 513 break;
506 514
507 default: 515 default:
@@ -516,7 +524,7 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
516 qp->remote_qpn = attr->dest_qp_num; 524 qp->remote_qpn = attr->dest_qp_num;
517 525
518 if (attr_mask & IB_QP_SQ_PSN) { 526 if (attr_mask & IB_QP_SQ_PSN) {
519 qp->s_next_psn = attr->sq_psn; 527 qp->s_psn = qp->s_next_psn = attr->sq_psn;
520 qp->s_last_psn = qp->s_next_psn - 1; 528 qp->s_last_psn = qp->s_next_psn - 1;
521 } 529 }
522 530
diff --git a/drivers/infiniband/hw/ipath/ipath_rc.c b/drivers/infiniband/hw/ipath/ipath_rc.c
index a08654042c03..a504cf67f272 100644
--- a/drivers/infiniband/hw/ipath/ipath_rc.c
+++ b/drivers/infiniband/hw/ipath/ipath_rc.c
@@ -201,6 +201,18 @@ int ipath_make_rc_req(struct ipath_qp *qp,
201 qp->s_rnr_timeout) 201 qp->s_rnr_timeout)
202 goto done; 202 goto done;
203 203
204 /* Limit the number of packets sent without an ACK. */
205 if (ipath_cmp24(qp->s_psn, qp->s_last_psn + IPATH_PSN_CREDIT) > 0) {
206 qp->s_wait_credit = 1;
207 dev->n_rc_stalls++;
208 spin_lock(&dev->pending_lock);
209 if (list_empty(&qp->timerwait))
210 list_add_tail(&qp->timerwait,
211 &dev->pending[dev->pending_index]);
212 spin_unlock(&dev->pending_lock);
213 goto done;
214 }
215
204 /* header size in 32-bit words LRH+BTH = (8+12)/4. */ 216 /* header size in 32-bit words LRH+BTH = (8+12)/4. */
205 hwords = 5; 217 hwords = 5;
206 bth0 = 0; 218 bth0 = 0;
@@ -221,7 +233,7 @@ int ipath_make_rc_req(struct ipath_qp *qp,
221 /* Check if send work queue is empty. */ 233 /* Check if send work queue is empty. */
222 if (qp->s_tail == qp->s_head) 234 if (qp->s_tail == qp->s_head)
223 goto done; 235 goto done;
224 qp->s_psn = wqe->psn = qp->s_next_psn; 236 wqe->psn = qp->s_next_psn;
225 newreq = 1; 237 newreq = 1;
226 } 238 }
227 /* 239 /*
@@ -393,12 +405,6 @@ int ipath_make_rc_req(struct ipath_qp *qp,
393 ss = &qp->s_sge; 405 ss = &qp->s_sge;
394 len = qp->s_len; 406 len = qp->s_len;
395 if (len > pmtu) { 407 if (len > pmtu) {
396 /*
397 * Request an ACK every 1/2 MB to avoid retransmit
398 * timeouts.
399 */
400 if (((wqe->length - len) % (512 * 1024)) == 0)
401 bth2 |= 1 << 31;
402 len = pmtu; 408 len = pmtu;
403 break; 409 break;
404 } 410 }
@@ -435,12 +441,6 @@ int ipath_make_rc_req(struct ipath_qp *qp,
435 ss = &qp->s_sge; 441 ss = &qp->s_sge;
436 len = qp->s_len; 442 len = qp->s_len;
437 if (len > pmtu) { 443 if (len > pmtu) {
438 /*
439 * Request an ACK every 1/2 MB to avoid retransmit
440 * timeouts.
441 */
442 if (((wqe->length - len) % (512 * 1024)) == 0)
443 bth2 |= 1 << 31;
444 len = pmtu; 444 len = pmtu;
445 break; 445 break;
446 } 446 }
@@ -498,6 +498,8 @@ int ipath_make_rc_req(struct ipath_qp *qp,
498 */ 498 */
499 goto done; 499 goto done;
500 } 500 }
501 if (ipath_cmp24(qp->s_psn, qp->s_last_psn + IPATH_PSN_CREDIT - 1) >= 0)
502 bth2 |= 1 << 31; /* Request ACK. */
501 qp->s_len -= len; 503 qp->s_len -= len;
502 qp->s_hdrwords = hwords; 504 qp->s_hdrwords = hwords;
503 qp->s_cur_sge = ss; 505 qp->s_cur_sge = ss;
@@ -737,6 +739,15 @@ bail:
737 return; 739 return;
738} 740}
739 741
742static inline void update_last_psn(struct ipath_qp *qp, u32 psn)
743{
744 if (qp->s_wait_credit) {
745 qp->s_wait_credit = 0;
746 tasklet_hi_schedule(&qp->s_task);
747 }
748 qp->s_last_psn = psn;
749}
750
740/** 751/**
741 * do_rc_ack - process an incoming RC ACK 752 * do_rc_ack - process an incoming RC ACK
742 * @qp: the QP the ACK came in on 753 * @qp: the QP the ACK came in on
@@ -805,7 +816,7 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode)
805 * The last valid PSN seen is the previous 816 * The last valid PSN seen is the previous
806 * request's. 817 * request's.
807 */ 818 */
808 qp->s_last_psn = wqe->psn - 1; 819 update_last_psn(qp, wqe->psn - 1);
809 /* Retry this request. */ 820 /* Retry this request. */
810 ipath_restart_rc(qp, wqe->psn, &wc); 821 ipath_restart_rc(qp, wqe->psn, &wc);
811 /* 822 /*
@@ -864,7 +875,7 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode)
864 ipath_get_credit(qp, aeth); 875 ipath_get_credit(qp, aeth);
865 qp->s_rnr_retry = qp->s_rnr_retry_cnt; 876 qp->s_rnr_retry = qp->s_rnr_retry_cnt;
866 qp->s_retry = qp->s_retry_cnt; 877 qp->s_retry = qp->s_retry_cnt;
867 qp->s_last_psn = psn; 878 update_last_psn(qp, psn);
868 ret = 1; 879 ret = 1;
869 goto bail; 880 goto bail;
870 881
@@ -883,7 +894,7 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode)
883 goto bail; 894 goto bail;
884 895
885 /* The last valid PSN is the previous PSN. */ 896 /* The last valid PSN is the previous PSN. */
886 qp->s_last_psn = psn - 1; 897 update_last_psn(qp, psn - 1);
887 898
888 dev->n_rc_resends += (int)qp->s_psn - (int)psn; 899 dev->n_rc_resends += (int)qp->s_psn - (int)psn;
889 900
@@ -898,7 +909,7 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode)
898 case 3: /* NAK */ 909 case 3: /* NAK */
899 /* The last valid PSN seen is the previous request's. */ 910 /* The last valid PSN seen is the previous request's. */
900 if (qp->s_last != qp->s_tail) 911 if (qp->s_last != qp->s_tail)
901 qp->s_last_psn = wqe->psn - 1; 912 update_last_psn(qp, wqe->psn - 1);
902 switch ((aeth >> IPATH_AETH_CREDIT_SHIFT) & 913 switch ((aeth >> IPATH_AETH_CREDIT_SHIFT) &
903 IPATH_AETH_CREDIT_MASK) { 914 IPATH_AETH_CREDIT_MASK) {
904 case 0: /* PSN sequence error */ 915 case 0: /* PSN sequence error */
@@ -1071,7 +1082,7 @@ static inline void ipath_rc_rcv_resp(struct ipath_ibdev *dev,
1071 * since we don't want s_sge modified. 1082 * since we don't want s_sge modified.
1072 */ 1083 */
1073 qp->s_len -= pmtu; 1084 qp->s_len -= pmtu;
1074 qp->s_last_psn = psn; 1085 update_last_psn(qp, psn);
1075 spin_unlock_irqrestore(&qp->s_lock, flags); 1086 spin_unlock_irqrestore(&qp->s_lock, flags);
1076 ipath_copy_sge(&qp->s_sge, data, pmtu); 1087 ipath_copy_sge(&qp->s_sge, data, pmtu);
1077 goto bail; 1088 goto bail;
@@ -1223,7 +1234,7 @@ static inline int ipath_rc_rcv_error(struct ipath_ibdev *dev,
1223 * Address range must be a subset of the original 1234 * Address range must be a subset of the original
1224 * request and start on pmtu boundaries. 1235 * request and start on pmtu boundaries.
1225 */ 1236 */
1226 ok = ipath_rkey_ok(dev, &qp->s_rdma_sge, 1237 ok = ipath_rkey_ok(qp, &qp->s_rdma_sge,
1227 qp->s_rdma_len, vaddr, rkey, 1238 qp->s_rdma_len, vaddr, rkey,
1228 IB_ACCESS_REMOTE_READ); 1239 IB_ACCESS_REMOTE_READ);
1229 if (unlikely(!ok)) { 1240 if (unlikely(!ok)) {
@@ -1282,6 +1293,14 @@ done:
1282 return 1; 1293 return 1;
1283} 1294}
1284 1295
1296static void ipath_rc_error(struct ipath_qp *qp, enum ib_wc_status err)
1297{
1298 spin_lock_irq(&qp->s_lock);
1299 qp->state = IB_QPS_ERR;
1300 ipath_error_qp(qp, err);
1301 spin_unlock_irq(&qp->s_lock);
1302}
1303
1285/** 1304/**
1286 * ipath_rc_rcv - process an incoming RC packet 1305 * ipath_rc_rcv - process an incoming RC packet
1287 * @dev: the device this packet came in on 1306 * @dev: the device this packet came in on
@@ -1309,6 +1328,10 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
1309 struct ib_reth *reth; 1328 struct ib_reth *reth;
1310 int header_in_data; 1329 int header_in_data;
1311 1330
1331 /* Validate the SLID. See Ch. 9.6.1.5 */
1332 if (unlikely(be16_to_cpu(hdr->lrh[3]) != qp->remote_ah_attr.dlid))
1333 goto done;
1334
1312 /* Check for GRH */ 1335 /* Check for GRH */
1313 if (!has_grh) { 1336 if (!has_grh) {
1314 ohdr = &hdr->u.oth; 1337 ohdr = &hdr->u.oth;
@@ -1370,8 +1393,7 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
1370 */ 1393 */
1371 if (qp->r_ack_state >= OP(COMPARE_SWAP)) 1394 if (qp->r_ack_state >= OP(COMPARE_SWAP))
1372 goto send_ack; 1395 goto send_ack;
1373 /* XXX Flush WQEs */ 1396 ipath_rc_error(qp, IB_WC_REM_INV_REQ_ERR);
1374 qp->state = IB_QPS_ERR;
1375 qp->r_ack_state = OP(SEND_ONLY); 1397 qp->r_ack_state = OP(SEND_ONLY);
1376 qp->r_nak_state = IB_NAK_INVALID_REQUEST; 1398 qp->r_nak_state = IB_NAK_INVALID_REQUEST;
1377 qp->r_ack_psn = qp->r_psn; 1399 qp->r_ack_psn = qp->r_psn;
@@ -1477,9 +1499,9 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
1477 goto nack_inv; 1499 goto nack_inv;
1478 ipath_copy_sge(&qp->r_sge, data, tlen); 1500 ipath_copy_sge(&qp->r_sge, data, tlen);
1479 qp->r_msn++; 1501 qp->r_msn++;
1480 if (opcode == OP(RDMA_WRITE_LAST) || 1502 if (!qp->r_wrid_valid)
1481 opcode == OP(RDMA_WRITE_ONLY))
1482 break; 1503 break;
1504 qp->r_wrid_valid = 0;
1483 wc.wr_id = qp->r_wr_id; 1505 wc.wr_id = qp->r_wr_id;
1484 wc.status = IB_WC_SUCCESS; 1506 wc.status = IB_WC_SUCCESS;
1485 wc.opcode = IB_WC_RECV; 1507 wc.opcode = IB_WC_RECV;
@@ -1517,7 +1539,7 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
1517 int ok; 1539 int ok;
1518 1540
1519 /* Check rkey & NAK */ 1541 /* Check rkey & NAK */
1520 ok = ipath_rkey_ok(dev, &qp->r_sge, 1542 ok = ipath_rkey_ok(qp, &qp->r_sge,
1521 qp->r_len, vaddr, rkey, 1543 qp->r_len, vaddr, rkey,
1522 IB_ACCESS_REMOTE_WRITE); 1544 IB_ACCESS_REMOTE_WRITE);
1523 if (unlikely(!ok)) 1545 if (unlikely(!ok))
@@ -1559,7 +1581,7 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
1559 int ok; 1581 int ok;
1560 1582
1561 /* Check rkey & NAK */ 1583 /* Check rkey & NAK */
1562 ok = ipath_rkey_ok(dev, &qp->s_rdma_sge, 1584 ok = ipath_rkey_ok(qp, &qp->s_rdma_sge,
1563 qp->s_rdma_len, vaddr, rkey, 1585 qp->s_rdma_len, vaddr, rkey,
1564 IB_ACCESS_REMOTE_READ); 1586 IB_ACCESS_REMOTE_READ);
1565 if (unlikely(!ok)) { 1587 if (unlikely(!ok)) {
@@ -1618,7 +1640,7 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
1618 goto nack_inv; 1640 goto nack_inv;
1619 rkey = be32_to_cpu(ateth->rkey); 1641 rkey = be32_to_cpu(ateth->rkey);
1620 /* Check rkey & NAK */ 1642 /* Check rkey & NAK */
1621 if (unlikely(!ipath_rkey_ok(dev, &qp->r_sge, 1643 if (unlikely(!ipath_rkey_ok(qp, &qp->r_sge,
1622 sizeof(u64), vaddr, rkey, 1644 sizeof(u64), vaddr, rkey,
1623 IB_ACCESS_REMOTE_ATOMIC))) 1645 IB_ACCESS_REMOTE_ATOMIC)))
1624 goto nack_acc; 1646 goto nack_acc;
@@ -1670,8 +1692,7 @@ nack_acc:
1670 * is pending though. 1692 * is pending though.
1671 */ 1693 */
1672 if (qp->r_ack_state < OP(COMPARE_SWAP)) { 1694 if (qp->r_ack_state < OP(COMPARE_SWAP)) {
1673 /* XXX Flush WQEs */ 1695 ipath_rc_error(qp, IB_WC_REM_ACCESS_ERR);
1674 qp->state = IB_QPS_ERR;
1675 qp->r_ack_state = OP(RDMA_WRITE_ONLY); 1696 qp->r_ack_state = OP(RDMA_WRITE_ONLY);
1676 qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR; 1697 qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR;
1677 qp->r_ack_psn = qp->r_psn; 1698 qp->r_ack_psn = qp->r_psn;
diff --git a/drivers/infiniband/hw/ipath/ipath_registers.h b/drivers/infiniband/hw/ipath/ipath_registers.h
index 6e23b3d632b8..dffc76016d3c 100644
--- a/drivers/infiniband/hw/ipath/ipath_registers.h
+++ b/drivers/infiniband/hw/ipath/ipath_registers.h
@@ -134,10 +134,24 @@
134#define INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT 40 134#define INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT 40
135#define INFINIPATH_HWE_RXEMEMPARITYERR_MASK 0x7FULL 135#define INFINIPATH_HWE_RXEMEMPARITYERR_MASK 0x7FULL
136#define INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT 44 136#define INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT 44
137#define INFINIPATH_HWE_RXDSYNCMEMPARITYERR 0x0000000400000000ULL
138#define INFINIPATH_HWE_MEMBISTFAILED 0x0040000000000000ULL
139#define INFINIPATH_HWE_IBCBUSTOSPCPARITYERR 0x4000000000000000ULL 137#define INFINIPATH_HWE_IBCBUSTOSPCPARITYERR 0x4000000000000000ULL
140#define INFINIPATH_HWE_IBCBUSFRSPCPARITYERR 0x8000000000000000ULL 138#define INFINIPATH_HWE_IBCBUSFRSPCPARITYERR 0x8000000000000000ULL
139/* txe mem parity errors (shift by INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT) */
140#define INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF 0x1ULL
141#define INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC 0x2ULL
142#define INFINIPATH_HWE_TXEMEMPARITYERR_PIOLAUNCHFIFO 0x4ULL
143/* rxe mem parity errors (shift by INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT) */
144#define INFINIPATH_HWE_RXEMEMPARITYERR_RCVBUF 0x01ULL
145#define INFINIPATH_HWE_RXEMEMPARITYERR_LOOKUPQ 0x02ULL
146#define INFINIPATH_HWE_RXEMEMPARITYERR_EAGERTID 0x04ULL
147#define INFINIPATH_HWE_RXEMEMPARITYERR_EXPTID 0x08ULL
148#define INFINIPATH_HWE_RXEMEMPARITYERR_FLAGBUF 0x10ULL
149#define INFINIPATH_HWE_RXEMEMPARITYERR_DATAINFO 0x20ULL
150#define INFINIPATH_HWE_RXEMEMPARITYERR_HDRINFO 0x40ULL
151/* waldo specific -- find the rest in ipath_6110.c */
152#define INFINIPATH_HWE_RXDSYNCMEMPARITYERR 0x0000000400000000ULL
153/* monty specific -- find the rest in ipath_6120.c */
154#define INFINIPATH_HWE_MEMBISTFAILED 0x0040000000000000ULL
141 155
142/* kr_hwdiagctrl bits */ 156/* kr_hwdiagctrl bits */
143#define INFINIPATH_DC_FORCETXEMEMPARITYERR_MASK 0xFULL 157#define INFINIPATH_DC_FORCETXEMEMPARITYERR_MASK 0xFULL
@@ -209,9 +223,9 @@
209 223
210/* combination link status states that we use with some frequency */ 224/* combination link status states that we use with some frequency */
211#define IPATH_IBSTATE_MASK ((INFINIPATH_IBCS_LINKTRAININGSTATE_MASK \ 225#define IPATH_IBSTATE_MASK ((INFINIPATH_IBCS_LINKTRAININGSTATE_MASK \
212 << INFINIPATH_IBCS_LINKSTATE_SHIFT) | \ 226 << INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) | \
213 (INFINIPATH_IBCS_LINKSTATE_MASK \ 227 (INFINIPATH_IBCS_LINKSTATE_MASK \
214 <<INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT)) 228 <<INFINIPATH_IBCS_LINKSTATE_SHIFT))
215#define IPATH_IBSTATE_INIT ((INFINIPATH_IBCS_L_STATE_INIT \ 229#define IPATH_IBSTATE_INIT ((INFINIPATH_IBCS_L_STATE_INIT \
216 << INFINIPATH_IBCS_LINKSTATE_SHIFT) | \ 230 << INFINIPATH_IBCS_LINKSTATE_SHIFT) | \
217 (INFINIPATH_IBCS_LT_STATE_LINKUP \ 231 (INFINIPATH_IBCS_LT_STATE_LINKUP \
@@ -302,6 +316,17 @@
302 316
303typedef u64 ipath_err_t; 317typedef u64 ipath_err_t;
304 318
319/* The following change with the type of device, so
320 * need to be part of the ipath_devdata struct, or
321 * we could have problems plugging in devices of
322 * different types (e.g. one HT, one PCIE)
323 * in one system, to be managed by one driver.
324 * On the other hand, this file is may also be included
325 * by other code, so leave the declarations here
326 * temporarily. Minor footprint issue if common-model
327 * linker used, none if C89+ linker used.
328 */
329
305/* mask of defined bits for various registers */ 330/* mask of defined bits for various registers */
306extern u64 infinipath_i_bitsextant; 331extern u64 infinipath_i_bitsextant;
307extern ipath_err_t infinipath_e_bitsextant, infinipath_hwe_bitsextant; 332extern ipath_err_t infinipath_e_bitsextant, infinipath_hwe_bitsextant;
@@ -310,13 +335,6 @@ extern ipath_err_t infinipath_e_bitsextant, infinipath_hwe_bitsextant;
310extern u32 infinipath_i_rcvavail_mask, infinipath_i_rcvurg_mask; 335extern u32 infinipath_i_rcvavail_mask, infinipath_i_rcvurg_mask;
311 336
312/* 337/*
313 * register bits for selecting i2c direction and values, used for I2C serial
314 * flash
315 */
316extern u16 ipath_gpio_sda_num, ipath_gpio_scl_num;
317extern u64 ipath_gpio_sda, ipath_gpio_scl;
318
319/*
320 * These are the infinipath general register numbers (not offsets). 338 * These are the infinipath general register numbers (not offsets).
321 * The kernel registers are used directly, those beyond the kernel 339 * The kernel registers are used directly, those beyond the kernel
322 * registers are calculated from one of the base registers. The use of 340 * registers are calculated from one of the base registers. The use of
diff --git a/drivers/infiniband/hw/ipath/ipath_ruc.c b/drivers/infiniband/hw/ipath/ipath_ruc.c
index 5c1da2d25e03..f7530512045d 100644
--- a/drivers/infiniband/hw/ipath/ipath_ruc.c
+++ b/drivers/infiniband/hw/ipath/ipath_ruc.c
@@ -108,7 +108,6 @@ void ipath_insert_rnr_queue(struct ipath_qp *qp)
108 108
109static int init_sge(struct ipath_qp *qp, struct ipath_rwqe *wqe) 109static int init_sge(struct ipath_qp *qp, struct ipath_rwqe *wqe)
110{ 110{
111 struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
112 int user = to_ipd(qp->ibqp.pd)->user; 111 int user = to_ipd(qp->ibqp.pd)->user;
113 int i, j, ret; 112 int i, j, ret;
114 struct ib_wc wc; 113 struct ib_wc wc;
@@ -119,8 +118,7 @@ static int init_sge(struct ipath_qp *qp, struct ipath_rwqe *wqe)
119 continue; 118 continue;
120 /* Check LKEY */ 119 /* Check LKEY */
121 if ((user && wqe->sg_list[i].lkey == 0) || 120 if ((user && wqe->sg_list[i].lkey == 0) ||
122 !ipath_lkey_ok(&dev->lk_table, 121 !ipath_lkey_ok(qp, &qp->r_sg_list[j], &wqe->sg_list[i],
123 &qp->r_sg_list[j], &wqe->sg_list[i],
124 IB_ACCESS_LOCAL_WRITE)) 122 IB_ACCESS_LOCAL_WRITE))
125 goto bad_lkey; 123 goto bad_lkey;
126 qp->r_len += wqe->sg_list[i].length; 124 qp->r_len += wqe->sg_list[i].length;
@@ -231,6 +229,7 @@ int ipath_get_rwqe(struct ipath_qp *qp, int wr_id_only)
231 } 229 }
232 } 230 }
233 spin_unlock_irqrestore(&rq->lock, flags); 231 spin_unlock_irqrestore(&rq->lock, flags);
232 qp->r_wrid_valid = 1;
234 233
235bail: 234bail:
236 return ret; 235 return ret;
@@ -326,7 +325,7 @@ again:
326 case IB_WR_RDMA_WRITE: 325 case IB_WR_RDMA_WRITE:
327 if (wqe->length == 0) 326 if (wqe->length == 0)
328 break; 327 break;
329 if (unlikely(!ipath_rkey_ok(dev, &qp->r_sge, wqe->length, 328 if (unlikely(!ipath_rkey_ok(qp, &qp->r_sge, wqe->length,
330 wqe->wr.wr.rdma.remote_addr, 329 wqe->wr.wr.rdma.remote_addr,
331 wqe->wr.wr.rdma.rkey, 330 wqe->wr.wr.rdma.rkey,
332 IB_ACCESS_REMOTE_WRITE))) { 331 IB_ACCESS_REMOTE_WRITE))) {
@@ -350,7 +349,7 @@ again:
350 break; 349 break;
351 350
352 case IB_WR_RDMA_READ: 351 case IB_WR_RDMA_READ:
353 if (unlikely(!ipath_rkey_ok(dev, &sqp->s_sge, wqe->length, 352 if (unlikely(!ipath_rkey_ok(qp, &sqp->s_sge, wqe->length,
354 wqe->wr.wr.rdma.remote_addr, 353 wqe->wr.wr.rdma.remote_addr,
355 wqe->wr.wr.rdma.rkey, 354 wqe->wr.wr.rdma.rkey,
356 IB_ACCESS_REMOTE_READ))) 355 IB_ACCESS_REMOTE_READ)))
@@ -365,7 +364,7 @@ again:
365 364
366 case IB_WR_ATOMIC_CMP_AND_SWP: 365 case IB_WR_ATOMIC_CMP_AND_SWP:
367 case IB_WR_ATOMIC_FETCH_AND_ADD: 366 case IB_WR_ATOMIC_FETCH_AND_ADD:
368 if (unlikely(!ipath_rkey_ok(dev, &qp->r_sge, sizeof(u64), 367 if (unlikely(!ipath_rkey_ok(qp, &qp->r_sge, sizeof(u64),
369 wqe->wr.wr.rdma.remote_addr, 368 wqe->wr.wr.rdma.remote_addr,
370 wqe->wr.wr.rdma.rkey, 369 wqe->wr.wr.rdma.rkey,
371 IB_ACCESS_REMOTE_ATOMIC))) 370 IB_ACCESS_REMOTE_ATOMIC)))
@@ -575,8 +574,7 @@ int ipath_post_ruc_send(struct ipath_qp *qp, struct ib_send_wr *wr)
575 } 574 }
576 if (wr->sg_list[i].length == 0) 575 if (wr->sg_list[i].length == 0)
577 continue; 576 continue;
578 if (!ipath_lkey_ok(&to_idev(qp->ibqp.device)->lk_table, 577 if (!ipath_lkey_ok(qp, &wqe->sg_list[j], &wr->sg_list[i],
579 &wqe->sg_list[j], &wr->sg_list[i],
580 acc)) { 578 acc)) {
581 spin_unlock_irqrestore(&qp->s_lock, flags); 579 spin_unlock_irqrestore(&qp->s_lock, flags);
582 ret = -EINVAL; 580 ret = -EINVAL;
diff --git a/drivers/infiniband/hw/ipath/ipath_srq.c b/drivers/infiniband/hw/ipath/ipath_srq.c
index 941e866d9517..94033503400c 100644
--- a/drivers/infiniband/hw/ipath/ipath_srq.c
+++ b/drivers/infiniband/hw/ipath/ipath_srq.c
@@ -104,11 +104,6 @@ struct ib_srq *ipath_create_srq(struct ib_pd *ibpd,
104 u32 sz; 104 u32 sz;
105 struct ib_srq *ret; 105 struct ib_srq *ret;
106 106
107 if (dev->n_srqs_allocated == ib_ipath_max_srqs) {
108 ret = ERR_PTR(-ENOMEM);
109 goto done;
110 }
111
112 if (srq_init_attr->attr.max_wr == 0) { 107 if (srq_init_attr->attr.max_wr == 0) {
113 ret = ERR_PTR(-EINVAL); 108 ret = ERR_PTR(-EINVAL);
114 goto done; 109 goto done;
@@ -180,10 +175,17 @@ struct ib_srq *ipath_create_srq(struct ib_pd *ibpd,
180 spin_lock_init(&srq->rq.lock); 175 spin_lock_init(&srq->rq.lock);
181 srq->rq.wq->head = 0; 176 srq->rq.wq->head = 0;
182 srq->rq.wq->tail = 0; 177 srq->rq.wq->tail = 0;
183 srq->rq.max_sge = srq_init_attr->attr.max_sge;
184 srq->limit = srq_init_attr->attr.srq_limit; 178 srq->limit = srq_init_attr->attr.srq_limit;
185 179
186 dev->n_srqs_allocated++; 180 spin_lock(&dev->n_srqs_lock);
181 if (dev->n_srqs_allocated == ib_ipath_max_srqs) {
182 spin_unlock(&dev->n_srqs_lock);
183 ret = ERR_PTR(-ENOMEM);
184 goto bail_wq;
185 }
186
187 dev->n_srqs_allocated++;
188 spin_unlock(&dev->n_srqs_lock);
187 189
188 ret = &srq->ibsrq; 190 ret = &srq->ibsrq;
189 goto done; 191 goto done;
@@ -351,8 +353,13 @@ int ipath_destroy_srq(struct ib_srq *ibsrq)
351 struct ipath_srq *srq = to_isrq(ibsrq); 353 struct ipath_srq *srq = to_isrq(ibsrq);
352 struct ipath_ibdev *dev = to_idev(ibsrq->device); 354 struct ipath_ibdev *dev = to_idev(ibsrq->device);
353 355
356 spin_lock(&dev->n_srqs_lock);
354 dev->n_srqs_allocated--; 357 dev->n_srqs_allocated--;
355 vfree(srq->rq.wq); 358 spin_unlock(&dev->n_srqs_lock);
359 if (srq->ip)
360 kref_put(&srq->ip->ref, ipath_release_mmap_info);
361 else
362 vfree(srq->rq.wq);
356 kfree(srq); 363 kfree(srq);
357 364
358 return 0; 365 return 0;
diff --git a/drivers/infiniband/hw/ipath/ipath_sysfs.c b/drivers/infiniband/hw/ipath/ipath_sysfs.c
index e299148c4b68..182de34f9f47 100644
--- a/drivers/infiniband/hw/ipath/ipath_sysfs.c
+++ b/drivers/infiniband/hw/ipath/ipath_sysfs.c
@@ -257,7 +257,7 @@ static ssize_t store_guid(struct device *dev,
257 struct ipath_devdata *dd = dev_get_drvdata(dev); 257 struct ipath_devdata *dd = dev_get_drvdata(dev);
258 ssize_t ret; 258 ssize_t ret;
259 unsigned short guid[8]; 259 unsigned short guid[8];
260 __be64 nguid; 260 __be64 new_guid;
261 u8 *ng; 261 u8 *ng;
262 int i; 262 int i;
263 263
@@ -266,7 +266,7 @@ static ssize_t store_guid(struct device *dev,
266 &guid[4], &guid[5], &guid[6], &guid[7]) != 8) 266 &guid[4], &guid[5], &guid[6], &guid[7]) != 8)
267 goto invalid; 267 goto invalid;
268 268
269 ng = (u8 *) &nguid; 269 ng = (u8 *) &new_guid;
270 270
271 for (i = 0; i < 8; i++) { 271 for (i = 0; i < 8; i++) {
272 if (guid[i] > 0xff) 272 if (guid[i] > 0xff)
@@ -274,7 +274,10 @@ static ssize_t store_guid(struct device *dev,
274 ng[i] = guid[i]; 274 ng[i] = guid[i];
275 } 275 }
276 276
277 dd->ipath_guid = nguid; 277 if (new_guid == 0)
278 goto invalid;
279
280 dd->ipath_guid = new_guid;
278 dd->ipath_nguid = 1; 281 dd->ipath_nguid = 1;
279 282
280 ret = strlen(buf); 283 ret = strlen(buf);
@@ -297,6 +300,16 @@ static ssize_t show_nguid(struct device *dev,
297 return scnprintf(buf, PAGE_SIZE, "%u\n", dd->ipath_nguid); 300 return scnprintf(buf, PAGE_SIZE, "%u\n", dd->ipath_nguid);
298} 301}
299 302
303static ssize_t show_nports(struct device *dev,
304 struct device_attribute *attr,
305 char *buf)
306{
307 struct ipath_devdata *dd = dev_get_drvdata(dev);
308
309 /* Return the number of user ports available. */
310 return scnprintf(buf, PAGE_SIZE, "%u\n", dd->ipath_cfgports - 1);
311}
312
300static ssize_t show_serial(struct device *dev, 313static ssize_t show_serial(struct device *dev,
301 struct device_attribute *attr, 314 struct device_attribute *attr,
302 char *buf) 315 char *buf)
@@ -608,6 +621,7 @@ static DEVICE_ATTR(mlid, S_IWUSR | S_IRUGO, show_mlid, store_mlid);
608static DEVICE_ATTR(mtu, S_IWUSR | S_IRUGO, show_mtu, store_mtu); 621static DEVICE_ATTR(mtu, S_IWUSR | S_IRUGO, show_mtu, store_mtu);
609static DEVICE_ATTR(enabled, S_IWUSR | S_IRUGO, show_enabled, store_enabled); 622static DEVICE_ATTR(enabled, S_IWUSR | S_IRUGO, show_enabled, store_enabled);
610static DEVICE_ATTR(nguid, S_IRUGO, show_nguid, NULL); 623static DEVICE_ATTR(nguid, S_IRUGO, show_nguid, NULL);
624static DEVICE_ATTR(nports, S_IRUGO, show_nports, NULL);
611static DEVICE_ATTR(reset, S_IWUSR, NULL, store_reset); 625static DEVICE_ATTR(reset, S_IWUSR, NULL, store_reset);
612static DEVICE_ATTR(serial, S_IRUGO, show_serial, NULL); 626static DEVICE_ATTR(serial, S_IRUGO, show_serial, NULL);
613static DEVICE_ATTR(status, S_IRUGO, show_status, NULL); 627static DEVICE_ATTR(status, S_IRUGO, show_status, NULL);
@@ -623,6 +637,7 @@ static struct attribute *dev_attributes[] = {
623 &dev_attr_mlid.attr, 637 &dev_attr_mlid.attr,
624 &dev_attr_mtu.attr, 638 &dev_attr_mtu.attr,
625 &dev_attr_nguid.attr, 639 &dev_attr_nguid.attr,
640 &dev_attr_nports.attr,
626 &dev_attr_serial.attr, 641 &dev_attr_serial.attr,
627 &dev_attr_status.attr, 642 &dev_attr_status.attr,
628 &dev_attr_status_str.attr, 643 &dev_attr_status_str.attr,
diff --git a/drivers/infiniband/hw/ipath/ipath_uc.c b/drivers/infiniband/hw/ipath/ipath_uc.c
index 0fd3cded16ba..e636cfd67a82 100644
--- a/drivers/infiniband/hw/ipath/ipath_uc.c
+++ b/drivers/infiniband/hw/ipath/ipath_uc.c
@@ -246,6 +246,10 @@ void ipath_uc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
246 struct ib_reth *reth; 246 struct ib_reth *reth;
247 int header_in_data; 247 int header_in_data;
248 248
249 /* Validate the SLID. See Ch. 9.6.1.5 */
250 if (unlikely(be16_to_cpu(hdr->lrh[3]) != qp->remote_ah_attr.dlid))
251 goto done;
252
249 /* Check for GRH */ 253 /* Check for GRH */
250 if (!has_grh) { 254 if (!has_grh) {
251 ohdr = &hdr->u.oth; 255 ohdr = &hdr->u.oth;
@@ -440,7 +444,7 @@ void ipath_uc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
440 int ok; 444 int ok;
441 445
442 /* Check rkey */ 446 /* Check rkey */
443 ok = ipath_rkey_ok(dev, &qp->r_sge, qp->r_len, 447 ok = ipath_rkey_ok(qp, &qp->r_sge, qp->r_len,
444 vaddr, rkey, 448 vaddr, rkey,
445 IB_ACCESS_REMOTE_WRITE); 449 IB_ACCESS_REMOTE_WRITE);
446 if (unlikely(!ok)) { 450 if (unlikely(!ok)) {
diff --git a/drivers/infiniband/hw/ipath/ipath_ud.c b/drivers/infiniband/hw/ipath/ipath_ud.c
index 6991d1d74e3c..49f1102af8b3 100644
--- a/drivers/infiniband/hw/ipath/ipath_ud.c
+++ b/drivers/infiniband/hw/ipath/ipath_ud.c
@@ -39,7 +39,6 @@
39static int init_sge(struct ipath_qp *qp, struct ipath_rwqe *wqe, 39static int init_sge(struct ipath_qp *qp, struct ipath_rwqe *wqe,
40 u32 *lengthp, struct ipath_sge_state *ss) 40 u32 *lengthp, struct ipath_sge_state *ss)
41{ 41{
42 struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
43 int user = to_ipd(qp->ibqp.pd)->user; 42 int user = to_ipd(qp->ibqp.pd)->user;
44 int i, j, ret; 43 int i, j, ret;
45 struct ib_wc wc; 44 struct ib_wc wc;
@@ -50,8 +49,7 @@ static int init_sge(struct ipath_qp *qp, struct ipath_rwqe *wqe,
50 continue; 49 continue;
51 /* Check LKEY */ 50 /* Check LKEY */
52 if ((user && wqe->sg_list[i].lkey == 0) || 51 if ((user && wqe->sg_list[i].lkey == 0) ||
53 !ipath_lkey_ok(&dev->lk_table, 52 !ipath_lkey_ok(qp, j ? &ss->sg_list[j - 1] : &ss->sge,
54 j ? &ss->sg_list[j - 1] : &ss->sge,
55 &wqe->sg_list[i], IB_ACCESS_LOCAL_WRITE)) 53 &wqe->sg_list[i], IB_ACCESS_LOCAL_WRITE))
56 goto bad_lkey; 54 goto bad_lkey;
57 *lengthp += wqe->sg_list[i].length; 55 *lengthp += wqe->sg_list[i].length;
@@ -343,7 +341,7 @@ int ipath_post_ud_send(struct ipath_qp *qp, struct ib_send_wr *wr)
343 341
344 if (wr->sg_list[i].length == 0) 342 if (wr->sg_list[i].length == 0)
345 continue; 343 continue;
346 if (!ipath_lkey_ok(&dev->lk_table, ss.num_sge ? 344 if (!ipath_lkey_ok(qp, ss.num_sge ?
347 sg_list + ss.num_sge - 1 : &ss.sge, 345 sg_list + ss.num_sge - 1 : &ss.sge,
348 &wr->sg_list[i], 0)) { 346 &wr->sg_list[i], 0)) {
349 ret = -EINVAL; 347 ret = -EINVAL;
diff --git a/drivers/infiniband/hw/ipath/ipath_user_pages.c b/drivers/infiniband/hw/ipath/ipath_user_pages.c
index e32fca9faf80..413754b1d8a2 100644
--- a/drivers/infiniband/hw/ipath/ipath_user_pages.c
+++ b/drivers/infiniband/hw/ipath/ipath_user_pages.c
@@ -90,6 +90,62 @@ bail:
90} 90}
91 91
92/** 92/**
93 * ipath_map_page - a safety wrapper around pci_map_page()
94 *
95 * A dma_addr of all 0's is interpreted by the chip as "disabled".
96 * Unfortunately, it can also be a valid dma_addr returned on some
97 * architectures.
98 *
99 * The powerpc iommu assigns dma_addrs in ascending order, so we don't
100 * have to bother with retries or mapping a dummy page to insure we
101 * don't just get the same mapping again.
102 *
103 * I'm sure we won't be so lucky with other iommu's, so FIXME.
104 */
105dma_addr_t ipath_map_page(struct pci_dev *hwdev, struct page *page,
106 unsigned long offset, size_t size, int direction)
107{
108 dma_addr_t phys;
109
110 phys = pci_map_page(hwdev, page, offset, size, direction);
111
112 if (phys == 0) {
113 pci_unmap_page(hwdev, phys, size, direction);
114 phys = pci_map_page(hwdev, page, offset, size, direction);
115 /*
116 * FIXME: If we get 0 again, we should keep this page,
117 * map another, then free the 0 page.
118 */
119 }
120
121 return phys;
122}
123
124/**
125 * ipath_map_single - a safety wrapper around pci_map_single()
126 *
127 * Same idea as ipath_map_page().
128 */
129dma_addr_t ipath_map_single(struct pci_dev *hwdev, void *ptr, size_t size,
130 int direction)
131{
132 dma_addr_t phys;
133
134 phys = pci_map_single(hwdev, ptr, size, direction);
135
136 if (phys == 0) {
137 pci_unmap_single(hwdev, phys, size, direction);
138 phys = pci_map_single(hwdev, ptr, size, direction);
139 /*
140 * FIXME: If we get 0 again, we should keep this page,
141 * map another, then free the 0 page.
142 */
143 }
144
145 return phys;
146}
147
148/**
93 * ipath_get_user_pages - lock user pages into memory 149 * ipath_get_user_pages - lock user pages into memory
94 * @start_page: the start page 150 * @start_page: the start page
95 * @num_pages: the number of pages 151 * @num_pages: the number of pages
diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c
index b8381c5e72bd..42eaed88c281 100644
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c
@@ -898,7 +898,8 @@ int ipath_get_counters(struct ipath_devdata *dd,
898 ipath_snap_cntr(dd, dd->ipath_cregs->cr_erricrccnt) + 898 ipath_snap_cntr(dd, dd->ipath_cregs->cr_erricrccnt) +
899 ipath_snap_cntr(dd, dd->ipath_cregs->cr_errvcrccnt) + 899 ipath_snap_cntr(dd, dd->ipath_cregs->cr_errvcrccnt) +
900 ipath_snap_cntr(dd, dd->ipath_cregs->cr_errlpcrccnt) + 900 ipath_snap_cntr(dd, dd->ipath_cregs->cr_errlpcrccnt) +
901 ipath_snap_cntr(dd, dd->ipath_cregs->cr_badformatcnt); 901 ipath_snap_cntr(dd, dd->ipath_cregs->cr_badformatcnt) +
902 dd->ipath_rxfc_unsupvl_errs;
902 cntrs->port_rcv_remphys_errors = 903 cntrs->port_rcv_remphys_errors =
903 ipath_snap_cntr(dd, dd->ipath_cregs->cr_rcvebpcnt); 904 ipath_snap_cntr(dd, dd->ipath_cregs->cr_rcvebpcnt);
904 cntrs->port_xmit_discards = 905 cntrs->port_xmit_discards =
@@ -911,8 +912,10 @@ int ipath_get_counters(struct ipath_devdata *dd,
911 ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt); 912 ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt);
912 cntrs->port_rcv_packets = 913 cntrs->port_rcv_packets =
913 ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt); 914 ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt);
914 cntrs->local_link_integrity_errors = dd->ipath_lli_errors; 915 cntrs->local_link_integrity_errors =
915 cntrs->excessive_buffer_overrun_errors = 0; /* XXX */ 916 (dd->ipath_flags & IPATH_GPIO_ERRINTRS) ?
917 dd->ipath_lli_errs : dd->ipath_lli_errors;
918 cntrs->excessive_buffer_overrun_errors = dd->ipath_overrun_thresh_errs;
916 919
917 ret = 0; 920 ret = 0;
918 921
@@ -1199,6 +1202,7 @@ static struct ib_ah *ipath_create_ah(struct ib_pd *pd,
1199 struct ipath_ah *ah; 1202 struct ipath_ah *ah;
1200 struct ib_ah *ret; 1203 struct ib_ah *ret;
1201 struct ipath_ibdev *dev = to_idev(pd->device); 1204 struct ipath_ibdev *dev = to_idev(pd->device);
1205 unsigned long flags;
1202 1206
1203 /* A multicast address requires a GRH (see ch. 8.4.1). */ 1207 /* A multicast address requires a GRH (see ch. 8.4.1). */
1204 if (ah_attr->dlid >= IPATH_MULTICAST_LID_BASE && 1208 if (ah_attr->dlid >= IPATH_MULTICAST_LID_BASE &&
@@ -1225,16 +1229,16 @@ static struct ib_ah *ipath_create_ah(struct ib_pd *pd,
1225 goto bail; 1229 goto bail;
1226 } 1230 }
1227 1231
1228 spin_lock(&dev->n_ahs_lock); 1232 spin_lock_irqsave(&dev->n_ahs_lock, flags);
1229 if (dev->n_ahs_allocated == ib_ipath_max_ahs) { 1233 if (dev->n_ahs_allocated == ib_ipath_max_ahs) {
1230 spin_unlock(&dev->n_ahs_lock); 1234 spin_unlock_irqrestore(&dev->n_ahs_lock, flags);
1231 kfree(ah); 1235 kfree(ah);
1232 ret = ERR_PTR(-ENOMEM); 1236 ret = ERR_PTR(-ENOMEM);
1233 goto bail; 1237 goto bail;
1234 } 1238 }
1235 1239
1236 dev->n_ahs_allocated++; 1240 dev->n_ahs_allocated++;
1237 spin_unlock(&dev->n_ahs_lock); 1241 spin_unlock_irqrestore(&dev->n_ahs_lock, flags);
1238 1242
1239 /* ib_create_ah() will initialize ah->ibah. */ 1243 /* ib_create_ah() will initialize ah->ibah. */
1240 ah->attr = *ah_attr; 1244 ah->attr = *ah_attr;
@@ -1255,10 +1259,11 @@ static int ipath_destroy_ah(struct ib_ah *ibah)
1255{ 1259{
1256 struct ipath_ibdev *dev = to_idev(ibah->device); 1260 struct ipath_ibdev *dev = to_idev(ibah->device);
1257 struct ipath_ah *ah = to_iah(ibah); 1261 struct ipath_ah *ah = to_iah(ibah);
1262 unsigned long flags;
1258 1263
1259 spin_lock(&dev->n_ahs_lock); 1264 spin_lock_irqsave(&dev->n_ahs_lock, flags);
1260 dev->n_ahs_allocated--; 1265 dev->n_ahs_allocated--;
1261 spin_unlock(&dev->n_ahs_lock); 1266 spin_unlock_irqrestore(&dev->n_ahs_lock, flags);
1262 1267
1263 kfree(ah); 1268 kfree(ah);
1264 1269
@@ -1380,11 +1385,13 @@ static int enable_timer(struct ipath_devdata *dd)
1380 * processing. 1385 * processing.
1381 */ 1386 */
1382 if (dd->ipath_flags & IPATH_GPIO_INTR) { 1387 if (dd->ipath_flags & IPATH_GPIO_INTR) {
1388 u64 val;
1383 ipath_write_kreg(dd, dd->ipath_kregs->kr_debugportselect, 1389 ipath_write_kreg(dd, dd->ipath_kregs->kr_debugportselect,
1384 0x2074076542310ULL); 1390 0x2074076542310ULL);
1385 /* Enable GPIO bit 2 interrupt */ 1391 /* Enable GPIO bit 2 interrupt */
1386 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask, 1392 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_gpio_mask);
1387 (u64) (1 << 2)); 1393 val |= (u64) (1 << IPATH_GPIO_PORT0_BIT);
1394 ipath_write_kreg( dd, dd->ipath_kregs->kr_gpio_mask, val);
1388 } 1395 }
1389 1396
1390 init_timer(&dd->verbs_timer); 1397 init_timer(&dd->verbs_timer);
@@ -1399,8 +1406,17 @@ static int enable_timer(struct ipath_devdata *dd)
1399static int disable_timer(struct ipath_devdata *dd) 1406static int disable_timer(struct ipath_devdata *dd)
1400{ 1407{
1401 /* Disable GPIO bit 2 interrupt */ 1408 /* Disable GPIO bit 2 interrupt */
1402 if (dd->ipath_flags & IPATH_GPIO_INTR) 1409 if (dd->ipath_flags & IPATH_GPIO_INTR) {
1403 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask, 0); 1410 u64 val;
1411 /* Disable GPIO bit 2 interrupt */
1412 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_gpio_mask);
1413 val &= ~((u64) (1 << IPATH_GPIO_PORT0_BIT));
1414 ipath_write_kreg( dd, dd->ipath_kregs->kr_gpio_mask, val);
1415 /*
1416 * We might want to undo changes to debugportselect,
1417 * but how?
1418 */
1419 }
1404 1420
1405 del_timer_sync(&dd->verbs_timer); 1421 del_timer_sync(&dd->verbs_timer);
1406 1422
@@ -1683,6 +1699,7 @@ static ssize_t show_stats(struct class_device *cdev, char *buf)
1683 "RC OTH NAKs %d\n" 1699 "RC OTH NAKs %d\n"
1684 "RC timeouts %d\n" 1700 "RC timeouts %d\n"
1685 "RC RDMA dup %d\n" 1701 "RC RDMA dup %d\n"
1702 "RC stalls %d\n"
1686 "piobuf wait %d\n" 1703 "piobuf wait %d\n"
1687 "no piobuf %d\n" 1704 "no piobuf %d\n"
1688 "PKT drops %d\n" 1705 "PKT drops %d\n"
@@ -1690,7 +1707,7 @@ static ssize_t show_stats(struct class_device *cdev, char *buf)
1690 dev->n_rc_resends, dev->n_rc_qacks, dev->n_rc_acks, 1707 dev->n_rc_resends, dev->n_rc_qacks, dev->n_rc_acks,
1691 dev->n_seq_naks, dev->n_rdma_seq, dev->n_rnr_naks, 1708 dev->n_seq_naks, dev->n_rdma_seq, dev->n_rnr_naks,
1692 dev->n_other_naks, dev->n_timeouts, 1709 dev->n_other_naks, dev->n_timeouts,
1693 dev->n_rdma_dup_busy, dev->n_piowait, 1710 dev->n_rdma_dup_busy, dev->n_rc_stalls, dev->n_piowait,
1694 dev->n_no_piobuf, dev->n_pkt_drops, dev->n_wqe_errs); 1711 dev->n_no_piobuf, dev->n_pkt_drops, dev->n_wqe_errs);
1695 for (i = 0; i < ARRAY_SIZE(dev->opstats); i++) { 1712 for (i = 0; i < ARRAY_SIZE(dev->opstats); i++) {
1696 const struct ipath_opcode_stats *si = &dev->opstats[i]; 1713 const struct ipath_opcode_stats *si = &dev->opstats[i];
diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h
index 09bbb3f9a217..8039f6e5f0c8 100644
--- a/drivers/infiniband/hw/ipath/ipath_verbs.h
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.h
@@ -220,6 +220,7 @@ struct ipath_segarray {
220}; 220};
221 221
222struct ipath_mregion { 222struct ipath_mregion {
223 struct ib_pd *pd; /* shares refcnt of ibmr.pd */
223 u64 user_base; /* User's address for this region */ 224 u64 user_base; /* User's address for this region */
224 u64 iova; /* IB start address of this region */ 225 u64 iova; /* IB start address of this region */
225 size_t length; 226 size_t length;
@@ -364,12 +365,14 @@ struct ipath_qp {
364 u8 r_min_rnr_timer; /* retry timeout value for RNR NAKs */ 365 u8 r_min_rnr_timer; /* retry timeout value for RNR NAKs */
365 u8 r_reuse_sge; /* for UC receive errors */ 366 u8 r_reuse_sge; /* for UC receive errors */
366 u8 r_sge_inx; /* current index into sg_list */ 367 u8 r_sge_inx; /* current index into sg_list */
368 u8 r_wrid_valid; /* r_wrid set but CQ entry not yet made */
367 u8 qp_access_flags; 369 u8 qp_access_flags;
368 u8 s_max_sge; /* size of s_wq->sg_list */ 370 u8 s_max_sge; /* size of s_wq->sg_list */
369 u8 s_retry_cnt; /* number of times to retry */ 371 u8 s_retry_cnt; /* number of times to retry */
370 u8 s_rnr_retry_cnt; 372 u8 s_rnr_retry_cnt;
371 u8 s_retry; /* requester retry counter */ 373 u8 s_retry; /* requester retry counter */
372 u8 s_rnr_retry; /* requester RNR retry counter */ 374 u8 s_rnr_retry; /* requester RNR retry counter */
375 u8 s_wait_credit; /* limit number of unacked packets sent */
373 u8 s_pkey_index; /* PKEY index to use */ 376 u8 s_pkey_index; /* PKEY index to use */
374 u8 timeout; /* Timeout for this QP */ 377 u8 timeout; /* Timeout for this QP */
375 enum ib_mtu path_mtu; 378 enum ib_mtu path_mtu;
@@ -393,6 +396,8 @@ struct ipath_qp {
393#define IPATH_S_BUSY 0 396#define IPATH_S_BUSY 0
394#define IPATH_S_SIGNAL_REQ_WR 1 397#define IPATH_S_SIGNAL_REQ_WR 1
395 398
399#define IPATH_PSN_CREDIT 2048
400
396/* 401/*
397 * Since struct ipath_swqe is not a fixed size, we can't simply index into 402 * Since struct ipath_swqe is not a fixed size, we can't simply index into
398 * struct ipath_qp.s_wq. This function does the array index computation. 403 * struct ipath_qp.s_wq. This function does the array index computation.
@@ -521,6 +526,7 @@ struct ipath_ibdev {
521 u32 n_rnr_naks; 526 u32 n_rnr_naks;
522 u32 n_other_naks; 527 u32 n_other_naks;
523 u32 n_timeouts; 528 u32 n_timeouts;
529 u32 n_rc_stalls;
524 u32 n_pkt_drops; 530 u32 n_pkt_drops;
525 u32 n_vl15_dropped; 531 u32 n_vl15_dropped;
526 u32 n_wqe_errs; 532 u32 n_wqe_errs;
@@ -634,6 +640,8 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
634 640
635int ipath_destroy_qp(struct ib_qp *ibqp); 641int ipath_destroy_qp(struct ib_qp *ibqp);
636 642
643void ipath_error_qp(struct ipath_qp *qp, enum ib_wc_status err);
644
637int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, 645int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
638 int attr_mask, struct ib_udata *udata); 646 int attr_mask, struct ib_udata *udata);
639 647
@@ -653,12 +661,6 @@ int ipath_verbs_send(struct ipath_devdata *dd, u32 hdrwords,
653 661
654void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int sig); 662void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int sig);
655 663
656int ipath_rkey_ok(struct ipath_ibdev *dev, struct ipath_sge_state *ss,
657 u32 len, u64 vaddr, u32 rkey, int acc);
658
659int ipath_lkey_ok(struct ipath_lkey_table *rkt, struct ipath_sge *isge,
660 struct ib_sge *sge, int acc);
661
662void ipath_copy_sge(struct ipath_sge_state *ss, void *data, u32 length); 664void ipath_copy_sge(struct ipath_sge_state *ss, void *data, u32 length);
663 665
664void ipath_skip_sge(struct ipath_sge_state *ss, u32 length); 666void ipath_skip_sge(struct ipath_sge_state *ss, u32 length);
@@ -683,10 +685,10 @@ int ipath_alloc_lkey(struct ipath_lkey_table *rkt,
683 685
684void ipath_free_lkey(struct ipath_lkey_table *rkt, u32 lkey); 686void ipath_free_lkey(struct ipath_lkey_table *rkt, u32 lkey);
685 687
686int ipath_lkey_ok(struct ipath_lkey_table *rkt, struct ipath_sge *isge, 688int ipath_lkey_ok(struct ipath_qp *qp, struct ipath_sge *isge,
687 struct ib_sge *sge, int acc); 689 struct ib_sge *sge, int acc);
688 690
689int ipath_rkey_ok(struct ipath_ibdev *dev, struct ipath_sge_state *ss, 691int ipath_rkey_ok(struct ipath_qp *qp, struct ipath_sge_state *ss,
690 u32 len, u64 vaddr, u32 rkey, int acc); 692 u32 len, u64 vaddr, u32 rkey, int acc);
691 693
692int ipath_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr, 694int ipath_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
diff --git a/drivers/infiniband/hw/ipath/ipath_wc_ppc64.c b/drivers/infiniband/hw/ipath/ipath_wc_ppc64.c
index 036fde662aa9..0095bb70f34e 100644
--- a/drivers/infiniband/hw/ipath/ipath_wc_ppc64.c
+++ b/drivers/infiniband/hw/ipath/ipath_wc_ppc64.c
@@ -38,13 +38,23 @@
38#include "ipath_kernel.h" 38#include "ipath_kernel.h"
39 39
40/** 40/**
41 * ipath_unordered_wc - indicate whether write combining is ordered 41 * ipath_enable_wc - enable write combining for MMIO writes to the device
42 * @dd: infinipath device
42 * 43 *
43 * PowerPC systems (at least those in the 970 processor family) 44 * Nothing to do on PowerPC, so just return without error.
44 * write partially filled store buffers in address order, but will write 45 */
45 * completely filled store buffers in "random" order, and therefore must 46int ipath_enable_wc(struct ipath_devdata *dd)
46 * have serialization for correctness with current InfiniPath chips. 47{
48 return 0;
49}
50
51/**
52 * ipath_unordered_wc - indicate whether write combining is unordered
47 * 53 *
54 * Because our performance depends on our ability to do write
55 * combining mmio writes in the most efficient way, we need to
56 * know if we are on a processor that may reorder stores when
57 * write combining.
48 */ 58 */
49int ipath_unordered_wc(void) 59int ipath_unordered_wc(void)
50{ 60{
diff --git a/drivers/infiniband/hw/ipath/ipath_wc_x86_64.c b/drivers/infiniband/hw/ipath/ipath_wc_x86_64.c
index f8f9e2e8cbdd..04696e62da87 100644
--- a/drivers/infiniband/hw/ipath/ipath_wc_x86_64.c
+++ b/drivers/infiniband/hw/ipath/ipath_wc_x86_64.c
@@ -123,6 +123,8 @@ int ipath_enable_wc(struct ipath_devdata *dd)
123 ipath_cdbg(VERBOSE, "Set mtrr for chip to WC, " 123 ipath_cdbg(VERBOSE, "Set mtrr for chip to WC, "
124 "cookie is %d\n", cookie); 124 "cookie is %d\n", cookie);
125 dd->ipath_wc_cookie = cookie; 125 dd->ipath_wc_cookie = cookie;
126 dd->ipath_wc_base = (unsigned long) pioaddr;
127 dd->ipath_wc_len = (unsigned long) piolen;
126 } 128 }
127 } 129 }
128 130
@@ -136,9 +138,16 @@ int ipath_enable_wc(struct ipath_devdata *dd)
136void ipath_disable_wc(struct ipath_devdata *dd) 138void ipath_disable_wc(struct ipath_devdata *dd)
137{ 139{
138 if (dd->ipath_wc_cookie) { 140 if (dd->ipath_wc_cookie) {
141 int r;
139 ipath_cdbg(VERBOSE, "undoing WCCOMB on pio buffers\n"); 142 ipath_cdbg(VERBOSE, "undoing WCCOMB on pio buffers\n");
140 mtrr_del(dd->ipath_wc_cookie, 0, 0); 143 r = mtrr_del(dd->ipath_wc_cookie, dd->ipath_wc_base,
141 dd->ipath_wc_cookie = 0; 144 dd->ipath_wc_len);
145 if (r < 0)
146 dev_info(&dd->pcidev->dev,
147 "mtrr_del(%lx, %lx, %lx) failed: %d\n",
148 dd->ipath_wc_cookie, dd->ipath_wc_base,
149 dd->ipath_wc_len, r);
150 dd->ipath_wc_cookie = 0; /* even on failure */
142 } 151 }
143} 152}
144 153
diff --git a/drivers/infiniband/ulp/iser/Kconfig b/drivers/infiniband/ulp/iser/Kconfig
index 365a1b5f19e0..aecbb9083f0c 100644
--- a/drivers/infiniband/ulp/iser/Kconfig
+++ b/drivers/infiniband/ulp/iser/Kconfig
@@ -1,11 +1,12 @@
1config INFINIBAND_ISER 1config INFINIBAND_ISER
2 tristate "ISCSI RDMA Protocol" 2 tristate "iSCSI Extensions for RDMA (iSER)"
3 depends on INFINIBAND && SCSI && INET 3 depends on INFINIBAND && SCSI && INET
4 select SCSI_ISCSI_ATTRS 4 select SCSI_ISCSI_ATTRS
5 ---help--- 5 ---help---
6 Support for the ISCSI RDMA Protocol over InfiniBand. This 6 Support for the iSCSI Extensions for RDMA (iSER) Protocol
7 allows you to access storage devices that speak ISER/ISCSI 7 over InfiniBand. This allows you to access storage devices
8 over InfiniBand. 8 that speak iSCSI over iSER over InfiniBand.
9 9
10 The ISER protocol is defined by IETF. 10 The iSER protocol is defined by IETF.
11 See <http://www.ietf.org/>. 11 See <http://www.ietf.org/internet-drafts/draft-ietf-ips-iser-05.txt>
12 and <http://www.infinibandta.org/members/spec/iser_annex_060418.pdf>
diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.c b/drivers/infiniband/ulp/iser/iscsi_iser.c
index 2a14fe2e3226..eb6f98d82289 100644
--- a/drivers/infiniband/ulp/iser/iscsi_iser.c
+++ b/drivers/infiniband/ulp/iser/iscsi_iser.c
@@ -317,6 +317,8 @@ iscsi_iser_conn_destroy(struct iscsi_cls_conn *cls_conn)
317 struct iscsi_iser_conn *iser_conn = conn->dd_data; 317 struct iscsi_iser_conn *iser_conn = conn->dd_data;
318 318
319 iscsi_conn_teardown(cls_conn); 319 iscsi_conn_teardown(cls_conn);
320 if (iser_conn->ib_conn)
321 iser_conn->ib_conn->iser_conn = NULL;
320 kfree(iser_conn); 322 kfree(iser_conn);
321} 323}
322 324
diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.h b/drivers/infiniband/ulp/iser/iscsi_iser.h
index 2cf9ae0def1c..9c53916f28c2 100644
--- a/drivers/infiniband/ulp/iser/iscsi_iser.h
+++ b/drivers/infiniband/ulp/iser/iscsi_iser.h
@@ -192,7 +192,7 @@ struct iser_regd_buf {
192 192
193struct iser_dto { 193struct iser_dto {
194 struct iscsi_iser_cmd_task *ctask; 194 struct iscsi_iser_cmd_task *ctask;
195 struct iscsi_iser_conn *conn; 195 struct iser_conn *ib_conn;
196 int notify_enable; 196 int notify_enable;
197 197
198 /* vector of registered buffers */ 198 /* vector of registered buffers */
@@ -355,4 +355,11 @@ int iser_post_send(struct iser_desc *tx_desc);
355 355
356int iser_conn_state_comp(struct iser_conn *ib_conn, 356int iser_conn_state_comp(struct iser_conn *ib_conn,
357 enum iser_ib_conn_state comp); 357 enum iser_ib_conn_state comp);
358
359int iser_dma_map_task_data(struct iscsi_iser_cmd_task *iser_ctask,
360 struct iser_data_buf *data,
361 enum iser_data_dir iser_dir,
362 enum dma_data_direction dma_dir);
363
364void iser_dma_unmap_task_data(struct iscsi_iser_cmd_task *iser_ctask);
358#endif 365#endif
diff --git a/drivers/infiniband/ulp/iser/iser_initiator.c b/drivers/infiniband/ulp/iser/iser_initiator.c
index ccf56f6f7236..9b3d79c796c8 100644
--- a/drivers/infiniband/ulp/iser/iser_initiator.c
+++ b/drivers/infiniband/ulp/iser/iser_initiator.c
@@ -66,42 +66,6 @@ static void iser_dto_add_regd_buff(struct iser_dto *dto,
66 dto->regd_vector_len++; 66 dto->regd_vector_len++;
67} 67}
68 68
69static int iser_dma_map_task_data(struct iscsi_iser_cmd_task *iser_ctask,
70 struct iser_data_buf *data,
71 enum iser_data_dir iser_dir,
72 enum dma_data_direction dma_dir)
73{
74 struct device *dma_device;
75
76 iser_ctask->dir[iser_dir] = 1;
77 dma_device = iser_ctask->iser_conn->ib_conn->device->ib_device->dma_device;
78
79 data->dma_nents = dma_map_sg(dma_device, data->buf, data->size, dma_dir);
80 if (data->dma_nents == 0) {
81 iser_err("dma_map_sg failed!!!\n");
82 return -EINVAL;
83 }
84 return 0;
85}
86
87static void iser_dma_unmap_task_data(struct iscsi_iser_cmd_task *iser_ctask)
88{
89 struct device *dma_device;
90 struct iser_data_buf *data;
91
92 dma_device = iser_ctask->iser_conn->ib_conn->device->ib_device->dma_device;
93
94 if (iser_ctask->dir[ISER_DIR_IN]) {
95 data = &iser_ctask->data[ISER_DIR_IN];
96 dma_unmap_sg(dma_device, data->buf, data->size, DMA_FROM_DEVICE);
97 }
98
99 if (iser_ctask->dir[ISER_DIR_OUT]) {
100 data = &iser_ctask->data[ISER_DIR_OUT];
101 dma_unmap_sg(dma_device, data->buf, data->size, DMA_TO_DEVICE);
102 }
103}
104
105/* Register user buffer memory and initialize passive rdma 69/* Register user buffer memory and initialize passive rdma
106 * dto descriptor. Total data size is stored in 70 * dto descriptor. Total data size is stored in
107 * iser_ctask->data[ISER_DIR_IN].data_len 71 * iser_ctask->data[ISER_DIR_IN].data_len
@@ -249,7 +213,7 @@ static int iser_post_receive_control(struct iscsi_conn *conn)
249 } 213 }
250 214
251 recv_dto = &rx_desc->dto; 215 recv_dto = &rx_desc->dto;
252 recv_dto->conn = iser_conn; 216 recv_dto->ib_conn = iser_conn->ib_conn;
253 recv_dto->regd_vector_len = 0; 217 recv_dto->regd_vector_len = 0;
254 218
255 regd_hdr = &rx_desc->hdr_regd_buf; 219 regd_hdr = &rx_desc->hdr_regd_buf;
@@ -296,7 +260,7 @@ static void iser_create_send_desc(struct iscsi_iser_conn *iser_conn,
296 regd_hdr->virt_addr = tx_desc; /* == &tx_desc->iser_header */ 260 regd_hdr->virt_addr = tx_desc; /* == &tx_desc->iser_header */
297 regd_hdr->data_size = ISER_TOTAL_HEADERS_LEN; 261 regd_hdr->data_size = ISER_TOTAL_HEADERS_LEN;
298 262
299 send_dto->conn = iser_conn; 263 send_dto->ib_conn = iser_conn->ib_conn;
300 send_dto->notify_enable = 1; 264 send_dto->notify_enable = 1;
301 send_dto->regd_vector_len = 0; 265 send_dto->regd_vector_len = 0;
302 266
@@ -588,7 +552,7 @@ void iser_rcv_completion(struct iser_desc *rx_desc,
588 unsigned long dto_xfer_len) 552 unsigned long dto_xfer_len)
589{ 553{
590 struct iser_dto *dto = &rx_desc->dto; 554 struct iser_dto *dto = &rx_desc->dto;
591 struct iscsi_iser_conn *conn = dto->conn; 555 struct iscsi_iser_conn *conn = dto->ib_conn->iser_conn;
592 struct iscsi_session *session = conn->iscsi_conn->session; 556 struct iscsi_session *session = conn->iscsi_conn->session;
593 struct iscsi_cmd_task *ctask; 557 struct iscsi_cmd_task *ctask;
594 struct iscsi_iser_cmd_task *iser_ctask; 558 struct iscsi_iser_cmd_task *iser_ctask;
@@ -641,7 +605,8 @@ void iser_rcv_completion(struct iser_desc *rx_desc,
641void iser_snd_completion(struct iser_desc *tx_desc) 605void iser_snd_completion(struct iser_desc *tx_desc)
642{ 606{
643 struct iser_dto *dto = &tx_desc->dto; 607 struct iser_dto *dto = &tx_desc->dto;
644 struct iscsi_iser_conn *iser_conn = dto->conn; 608 struct iser_conn *ib_conn = dto->ib_conn;
609 struct iscsi_iser_conn *iser_conn = ib_conn->iser_conn;
645 struct iscsi_conn *conn = iser_conn->iscsi_conn; 610 struct iscsi_conn *conn = iser_conn->iscsi_conn;
646 struct iscsi_mgmt_task *mtask; 611 struct iscsi_mgmt_task *mtask;
647 612
@@ -652,7 +617,7 @@ void iser_snd_completion(struct iser_desc *tx_desc)
652 if (tx_desc->type == ISCSI_TX_DATAOUT) 617 if (tx_desc->type == ISCSI_TX_DATAOUT)
653 kmem_cache_free(ig.desc_cache, tx_desc); 618 kmem_cache_free(ig.desc_cache, tx_desc);
654 619
655 atomic_dec(&iser_conn->ib_conn->post_send_buf_count); 620 atomic_dec(&ib_conn->post_send_buf_count);
656 621
657 write_lock(conn->recv_lock); 622 write_lock(conn->recv_lock);
658 if (conn->suspend_tx) { 623 if (conn->suspend_tx) {
@@ -698,14 +663,19 @@ void iser_ctask_rdma_init(struct iscsi_iser_cmd_task *iser_ctask)
698void iser_ctask_rdma_finalize(struct iscsi_iser_cmd_task *iser_ctask) 663void iser_ctask_rdma_finalize(struct iscsi_iser_cmd_task *iser_ctask)
699{ 664{
700 int deferred; 665 int deferred;
666 int is_rdma_aligned = 1;
701 667
702 /* if we were reading, copy back to unaligned sglist, 668 /* if we were reading, copy back to unaligned sglist,
703 * anyway dma_unmap and free the copy 669 * anyway dma_unmap and free the copy
704 */ 670 */
705 if (iser_ctask->data_copy[ISER_DIR_IN].copy_buf != NULL) 671 if (iser_ctask->data_copy[ISER_DIR_IN].copy_buf != NULL) {
672 is_rdma_aligned = 0;
706 iser_finalize_rdma_unaligned_sg(iser_ctask, ISER_DIR_IN); 673 iser_finalize_rdma_unaligned_sg(iser_ctask, ISER_DIR_IN);
707 if (iser_ctask->data_copy[ISER_DIR_OUT].copy_buf != NULL) 674 }
675 if (iser_ctask->data_copy[ISER_DIR_OUT].copy_buf != NULL) {
676 is_rdma_aligned = 0;
708 iser_finalize_rdma_unaligned_sg(iser_ctask, ISER_DIR_OUT); 677 iser_finalize_rdma_unaligned_sg(iser_ctask, ISER_DIR_OUT);
678 }
709 679
710 if (iser_ctask->dir[ISER_DIR_IN]) { 680 if (iser_ctask->dir[ISER_DIR_IN]) {
711 deferred = iser_regd_buff_release 681 deferred = iser_regd_buff_release
@@ -725,7 +695,9 @@ void iser_ctask_rdma_finalize(struct iscsi_iser_cmd_task *iser_ctask)
725 } 695 }
726 } 696 }
727 697
728 iser_dma_unmap_task_data(iser_ctask); 698 /* if the data was unaligned, it was already unmapped and then copied */
699 if (is_rdma_aligned)
700 iser_dma_unmap_task_data(iser_ctask);
729} 701}
730 702
731void iser_dto_buffs_release(struct iser_dto *dto) 703void iser_dto_buffs_release(struct iser_dto *dto)
diff --git a/drivers/infiniband/ulp/iser/iser_memory.c b/drivers/infiniband/ulp/iser/iser_memory.c
index d0b03f426581..0606744c3f84 100644
--- a/drivers/infiniband/ulp/iser/iser_memory.c
+++ b/drivers/infiniband/ulp/iser/iser_memory.c
@@ -369,6 +369,44 @@ static void iser_page_vec_build(struct iser_data_buf *data,
369 } 369 }
370} 370}
371 371
372int iser_dma_map_task_data(struct iscsi_iser_cmd_task *iser_ctask,
373 struct iser_data_buf *data,
374 enum iser_data_dir iser_dir,
375 enum dma_data_direction dma_dir)
376{
377 struct device *dma_device;
378
379 iser_ctask->dir[iser_dir] = 1;
380 dma_device =
381 iser_ctask->iser_conn->ib_conn->device->ib_device->dma_device;
382
383 data->dma_nents = dma_map_sg(dma_device, data->buf, data->size, dma_dir);
384 if (data->dma_nents == 0) {
385 iser_err("dma_map_sg failed!!!\n");
386 return -EINVAL;
387 }
388 return 0;
389}
390
391void iser_dma_unmap_task_data(struct iscsi_iser_cmd_task *iser_ctask)
392{
393 struct device *dma_device;
394 struct iser_data_buf *data;
395
396 dma_device =
397 iser_ctask->iser_conn->ib_conn->device->ib_device->dma_device;
398
399 if (iser_ctask->dir[ISER_DIR_IN]) {
400 data = &iser_ctask->data[ISER_DIR_IN];
401 dma_unmap_sg(dma_device, data->buf, data->size, DMA_FROM_DEVICE);
402 }
403
404 if (iser_ctask->dir[ISER_DIR_OUT]) {
405 data = &iser_ctask->data[ISER_DIR_OUT];
406 dma_unmap_sg(dma_device, data->buf, data->size, DMA_TO_DEVICE);
407 }
408}
409
372/** 410/**
373 * iser_reg_rdma_mem - Registers memory intended for RDMA, 411 * iser_reg_rdma_mem - Registers memory intended for RDMA,
374 * obtaining rkey and va 412 * obtaining rkey and va
@@ -394,6 +432,10 @@ int iser_reg_rdma_mem(struct iscsi_iser_cmd_task *iser_ctask,
394 iser_err("rdma alignment violation %d/%d aligned\n", 432 iser_err("rdma alignment violation %d/%d aligned\n",
395 aligned_len, mem->size); 433 aligned_len, mem->size);
396 iser_data_buf_dump(mem); 434 iser_data_buf_dump(mem);
435
436 /* unmap the command data before accessing it */
437 iser_dma_unmap_task_data(iser_ctask);
438
397 /* allocate copy buf, if we are writing, copy the */ 439 /* allocate copy buf, if we are writing, copy the */
398 /* unaligned scatterlist, dma map the copy */ 440 /* unaligned scatterlist, dma map the copy */
399 if (iser_start_rdma_unaligned_sg(iser_ctask, cmd_dir) != 0) 441 if (iser_start_rdma_unaligned_sg(iser_ctask, cmd_dir) != 0)
diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c
index ecdca7fc1e4c..18a000034996 100644
--- a/drivers/infiniband/ulp/iser/iser_verbs.c
+++ b/drivers/infiniband/ulp/iser/iser_verbs.c
@@ -571,6 +571,8 @@ void iser_conn_release(struct iser_conn *ib_conn)
571 /* on EVENT_ADDR_ERROR there's no device yet for this conn */ 571 /* on EVENT_ADDR_ERROR there's no device yet for this conn */
572 if (device != NULL) 572 if (device != NULL)
573 iser_device_try_release(device); 573 iser_device_try_release(device);
574 if (ib_conn->iser_conn)
575 ib_conn->iser_conn->ib_conn = NULL;
574 kfree(ib_conn); 576 kfree(ib_conn);
575} 577}
576 578
@@ -694,7 +696,7 @@ int iser_post_recv(struct iser_desc *rx_desc)
694 struct iser_dto *recv_dto = &rx_desc->dto; 696 struct iser_dto *recv_dto = &rx_desc->dto;
695 697
696 /* Retrieve conn */ 698 /* Retrieve conn */
697 ib_conn = recv_dto->conn->ib_conn; 699 ib_conn = recv_dto->ib_conn;
698 700
699 iser_dto_to_iov(recv_dto, iov, 2); 701 iser_dto_to_iov(recv_dto, iov, 2);
700 702
@@ -727,7 +729,7 @@ int iser_post_send(struct iser_desc *tx_desc)
727 struct iser_conn *ib_conn; 729 struct iser_conn *ib_conn;
728 struct iser_dto *dto = &tx_desc->dto; 730 struct iser_dto *dto = &tx_desc->dto;
729 731
730 ib_conn = dto->conn->ib_conn; 732 ib_conn = dto->ib_conn;
731 733
732 iser_dto_to_iov(dto, iov, MAX_REGD_BUF_VECTOR_LEN); 734 iser_dto_to_iov(dto, iov, MAX_REGD_BUF_VECTOR_LEN);
733 735
@@ -774,7 +776,7 @@ static void iser_comp_error_worker(void *data)
774static void iser_handle_comp_error(struct iser_desc *desc) 776static void iser_handle_comp_error(struct iser_desc *desc)
775{ 777{
776 struct iser_dto *dto = &desc->dto; 778 struct iser_dto *dto = &desc->dto;
777 struct iser_conn *ib_conn = dto->conn->ib_conn; 779 struct iser_conn *ib_conn = dto->ib_conn;
778 780
779 iser_dto_buffs_release(dto); 781 iser_dto_buffs_release(dto);
780 782
generated by cgit v1.2.2 (git 2.25.0) at 2025-10-23 20:13:50 -0400