1 /*
2  * This file is part of the Chelsio FCoE driver for Linux.
3  *
4  * Copyright (c) 2008-2012 Chelsio Communications, Inc. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 
37 #include <linux/kernel.h>
38 #include <linux/module.h>
39 #include <linux/init.h>
40 #include <linux/pci.h>
41 #include <linux/aer.h>
42 #include <linux/mm.h>
43 #include <linux/notifier.h>
44 #include <linux/kdebug.h>
45 #include <linux/seq_file.h>
46 #include <linux/debugfs.h>
47 #include <linux/string.h>
48 #include <linux/export.h>
49 
50 #include "csio_init.h"
51 #include "csio_defs.h"
52 
53 #define CSIO_MIN_MEMPOOL_SZ	64
54 
55 static struct dentry *csio_debugfs_root;
56 
57 static struct scsi_transport_template *csio_fcoe_transport;
58 static struct scsi_transport_template *csio_fcoe_transport_vport;
59 
60 /*
61  * debugfs support
62  */
63 static ssize_t
64 csio_mem_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
65 {
66 	loff_t pos = *ppos;
67 	loff_t avail = file_inode(file)->i_size;
68 	unsigned int mem = (uintptr_t)file->private_data & 3;
69 	struct csio_hw *hw = file->private_data - mem;
70 
71 	if (pos < 0)
72 		return -EINVAL;
73 	if (pos >= avail)
74 		return 0;
75 	if (count > avail - pos)
76 		count = avail - pos;
77 
78 	while (count) {
79 		size_t len;
80 		int ret, ofst;
81 		__be32 data[16];
82 
83 		if (mem == MEM_MC)
84 			ret = hw->chip_ops->chip_mc_read(hw, 0, pos,
85 							 data, NULL);
86 		else
87 			ret = hw->chip_ops->chip_edc_read(hw, mem, pos,
88 							  data, NULL);
89 		if (ret)
90 			return ret;
91 
92 		ofst = pos % sizeof(data);
93 		len = min(count, sizeof(data) - ofst);
94 		if (copy_to_user(buf, (u8 *)data + ofst, len))
95 			return -EFAULT;
96 
97 		buf += len;
98 		pos += len;
99 		count -= len;
100 	}
101 	count = pos - *ppos;
102 	*ppos = pos;
103 	return count;
104 }
105 
106 static const struct file_operations csio_mem_debugfs_fops = {
107 	.owner   = THIS_MODULE,
108 	.open    = simple_open,
109 	.read    = csio_mem_read,
110 	.llseek  = default_llseek,
111 };
112 
113 void csio_add_debugfs_mem(struct csio_hw *hw, const char *name,
114 				 unsigned int idx, unsigned int size_mb)
115 {
116 	debugfs_create_file_size(name, S_IRUSR, hw->debugfs_root,
117 				 (void *)hw + idx, &csio_mem_debugfs_fops,
118 				 size_mb << 20);
119 }
120 
121 static int csio_setup_debugfs(struct csio_hw *hw)
122 {
123 	int i;
124 
125 	if (IS_ERR_OR_NULL(hw->debugfs_root))
126 		return -1;
127 
128 	i = csio_rd_reg32(hw, MA_TARGET_MEM_ENABLE_A);
129 	if (i & EDRAM0_ENABLE_F)
130 		csio_add_debugfs_mem(hw, "edc0", MEM_EDC0, 5);
131 	if (i & EDRAM1_ENABLE_F)
132 		csio_add_debugfs_mem(hw, "edc1", MEM_EDC1, 5);
133 
134 	hw->chip_ops->chip_dfs_create_ext_mem(hw);
135 	return 0;
136 }
137 
138 /*
139  * csio_dfs_create - Creates and sets up per-hw debugfs.
140  *
141  */
142 static int
143 csio_dfs_create(struct csio_hw *hw)
144 {
145 	if (csio_debugfs_root) {
146 		hw->debugfs_root = debugfs_create_dir(pci_name(hw->pdev),
147 							csio_debugfs_root);
148 		csio_setup_debugfs(hw);
149 	}
150 
151 	return 0;
152 }
153 
154 /*
155  * csio_dfs_destroy - Destroys per-hw debugfs.
156  */
157 static int
158 csio_dfs_destroy(struct csio_hw *hw)
159 {
160 	if (hw->debugfs_root)
161 		debugfs_remove_recursive(hw->debugfs_root);
162 
163 	return 0;
164 }
165 
166 /*
167  * csio_dfs_init - Debug filesystem initialization for the module.
168  *
169  */
170 static int
171 csio_dfs_init(void)
172 {
173 	csio_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
174 	if (!csio_debugfs_root)
175 		pr_warn("Could not create debugfs entry, continuing\n");
176 
177 	return 0;
178 }
179 
180 /*
181  * csio_dfs_exit - debugfs cleanup for the module.
182  */
183 static void
184 csio_dfs_exit(void)
185 {
186 	debugfs_remove(csio_debugfs_root);
187 }
188 
189 /*
190  * csio_pci_init - PCI initialization.
191  * @pdev: PCI device.
192  * @bars: Bitmask of bars to be requested.
193  *
194  * Initializes the PCI function by enabling MMIO, setting bus
195  * mastership and setting DMA mask.
196  */
197 static int
198 csio_pci_init(struct pci_dev *pdev, int *bars)
199 {
200 	int rv = -ENODEV;
201 
202 	*bars = pci_select_bars(pdev, IORESOURCE_MEM);
203 
204 	if (pci_enable_device_mem(pdev))
205 		goto err;
206 
207 	if (pci_request_selected_regions(pdev, *bars, KBUILD_MODNAME))
208 		goto err_disable_device;
209 
210 	pci_set_master(pdev);
211 	pci_try_set_mwi(pdev);
212 
213 	if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
214 		pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
215 	} else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32))) {
216 		pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
217 	} else {
218 		dev_err(&pdev->dev, "No suitable DMA available.\n");
219 		goto err_release_regions;
220 	}
221 
222 	return 0;
223 
224 err_release_regions:
225 	pci_release_selected_regions(pdev, *bars);
226 err_disable_device:
227 	pci_disable_device(pdev);
228 err:
229 	return rv;
230 
231 }
232 
233 /*
234  * csio_pci_exit - PCI unitialization.
235  * @pdev: PCI device.
236  * @bars: Bars to be released.
237  *
238  */
239 static void
240 csio_pci_exit(struct pci_dev *pdev, int *bars)
241 {
242 	pci_release_selected_regions(pdev, *bars);
243 	pci_disable_device(pdev);
244 }
245 
246 /*
247  * csio_hw_init_workers - Initialize the HW module's worker threads.
248  * @hw: HW module.
249  *
250  */
251 static void
252 csio_hw_init_workers(struct csio_hw *hw)
253 {
254 	INIT_WORK(&hw->evtq_work, csio_evtq_worker);
255 }
256 
257 static void
258 csio_hw_exit_workers(struct csio_hw *hw)
259 {
260 	cancel_work_sync(&hw->evtq_work);
261 	flush_scheduled_work();
262 }
263 
264 static int
265 csio_create_queues(struct csio_hw *hw)
266 {
267 	int i, j;
268 	struct csio_mgmtm *mgmtm = csio_hw_to_mgmtm(hw);
269 	int rv;
270 	struct csio_scsi_cpu_info *info;
271 
272 	if (hw->flags & CSIO_HWF_Q_FW_ALLOCED)
273 		return 0;
274 
275 	if (hw->intr_mode != CSIO_IM_MSIX) {
276 		rv = csio_wr_iq_create(hw, NULL, hw->intr_iq_idx,
277 					0, hw->pport[0].portid, false, NULL);
278 		if (rv != 0) {
279 			csio_err(hw, " Forward Interrupt IQ failed!: %d\n", rv);
280 			return rv;
281 		}
282 	}
283 
284 	/* FW event queue */
285 	rv = csio_wr_iq_create(hw, NULL, hw->fwevt_iq_idx,
286 			       csio_get_fwevt_intr_idx(hw),
287 			       hw->pport[0].portid, true, NULL);
288 	if (rv != 0) {
289 		csio_err(hw, "FW event IQ config failed!: %d\n", rv);
290 		return rv;
291 	}
292 
293 	/* Create mgmt queue */
294 	rv = csio_wr_eq_create(hw, NULL, mgmtm->eq_idx,
295 			mgmtm->iq_idx, hw->pport[0].portid, NULL);
296 
297 	if (rv != 0) {
298 		csio_err(hw, "Mgmt EQ create failed!: %d\n", rv);
299 		goto err;
300 	}
301 
302 	/* Create SCSI queues */
303 	for (i = 0; i < hw->num_pports; i++) {
304 		info = &hw->scsi_cpu_info[i];
305 
306 		for (j = 0; j < info->max_cpus; j++) {
307 			struct csio_scsi_qset *sqset = &hw->sqset[i][j];
308 
309 			rv = csio_wr_iq_create(hw, NULL, sqset->iq_idx,
310 					       sqset->intr_idx, i, false, NULL);
311 			if (rv != 0) {
312 				csio_err(hw,
313 				   "SCSI module IQ config failed [%d][%d]:%d\n",
314 				   i, j, rv);
315 				goto err;
316 			}
317 			rv = csio_wr_eq_create(hw, NULL, sqset->eq_idx,
318 					       sqset->iq_idx, i, NULL);
319 			if (rv != 0) {
320 				csio_err(hw,
321 				   "SCSI module EQ config failed [%d][%d]:%d\n",
322 				   i, j, rv);
323 				goto err;
324 			}
325 		} /* for all CPUs */
326 	} /* For all ports */
327 
328 	hw->flags |= CSIO_HWF_Q_FW_ALLOCED;
329 	return 0;
330 err:
331 	csio_wr_destroy_queues(hw, true);
332 	return -EINVAL;
333 }
334 
335 /*
336  * csio_config_queues - Configure the DMA queues.
337  * @hw: HW module.
338  *
339  * Allocates memory for queues are registers them with FW.
340  */
341 int
342 csio_config_queues(struct csio_hw *hw)
343 {
344 	int i, j, idx, k = 0;
345 	int rv;
346 	struct csio_scsi_qset *sqset;
347 	struct csio_mgmtm *mgmtm = csio_hw_to_mgmtm(hw);
348 	struct csio_scsi_qset *orig;
349 	struct csio_scsi_cpu_info *info;
350 
351 	if (hw->flags & CSIO_HWF_Q_MEM_ALLOCED)
352 		return csio_create_queues(hw);
353 
354 	/* Calculate number of SCSI queues for MSIX we would like */
355 	hw->num_scsi_msix_cpus = num_online_cpus();
356 	hw->num_sqsets = num_online_cpus() * hw->num_pports;
357 
358 	if (hw->num_sqsets > CSIO_MAX_SCSI_QSETS) {
359 		hw->num_sqsets = CSIO_MAX_SCSI_QSETS;
360 		hw->num_scsi_msix_cpus = CSIO_MAX_SCSI_CPU;
361 	}
362 
363 	/* Initialize max_cpus, may get reduced during msix allocations */
364 	for (i = 0; i < hw->num_pports; i++)
365 		hw->scsi_cpu_info[i].max_cpus = hw->num_scsi_msix_cpus;
366 
367 	csio_dbg(hw, "nsqsets:%d scpus:%d\n",
368 		    hw->num_sqsets, hw->num_scsi_msix_cpus);
369 
370 	csio_intr_enable(hw);
371 
372 	if (hw->intr_mode != CSIO_IM_MSIX) {
373 
374 		/* Allocate Forward interrupt iq. */
375 		hw->intr_iq_idx = csio_wr_alloc_q(hw, CSIO_INTR_IQSIZE,
376 						CSIO_INTR_WRSIZE, CSIO_INGRESS,
377 						(void *)hw, 0, 0, NULL);
378 		if (hw->intr_iq_idx == -1) {
379 			csio_err(hw,
380 				 "Forward interrupt queue creation failed\n");
381 			goto intr_disable;
382 		}
383 	}
384 
385 	/* Allocate the FW evt queue */
386 	hw->fwevt_iq_idx = csio_wr_alloc_q(hw, CSIO_FWEVT_IQSIZE,
387 					   CSIO_FWEVT_WRSIZE,
388 					   CSIO_INGRESS, (void *)hw,
389 					   CSIO_FWEVT_FLBUFS, 0,
390 					   csio_fwevt_intx_handler);
391 	if (hw->fwevt_iq_idx == -1) {
392 		csio_err(hw, "FW evt queue creation failed\n");
393 		goto intr_disable;
394 	}
395 
396 	/* Allocate the mgmt queue */
397 	mgmtm->eq_idx = csio_wr_alloc_q(hw, CSIO_MGMT_EQSIZE,
398 				      CSIO_MGMT_EQ_WRSIZE,
399 				      CSIO_EGRESS, (void *)hw, 0, 0, NULL);
400 	if (mgmtm->eq_idx == -1) {
401 		csio_err(hw, "Failed to alloc egress queue for mgmt module\n");
402 		goto intr_disable;
403 	}
404 
405 	/* Use FW IQ for MGMT req completion */
406 	mgmtm->iq_idx = hw->fwevt_iq_idx;
407 
408 	/* Allocate SCSI queues */
409 	for (i = 0; i < hw->num_pports; i++) {
410 		info = &hw->scsi_cpu_info[i];
411 
412 		for (j = 0; j < hw->num_scsi_msix_cpus; j++) {
413 			sqset = &hw->sqset[i][j];
414 
415 			if (j >= info->max_cpus) {
416 				k = j % info->max_cpus;
417 				orig = &hw->sqset[i][k];
418 				sqset->eq_idx = orig->eq_idx;
419 				sqset->iq_idx = orig->iq_idx;
420 				continue;
421 			}
422 
423 			idx = csio_wr_alloc_q(hw, csio_scsi_eqsize, 0,
424 					      CSIO_EGRESS, (void *)hw, 0, 0,
425 					      NULL);
426 			if (idx == -1) {
427 				csio_err(hw, "EQ creation failed for idx:%d\n",
428 					    idx);
429 				goto intr_disable;
430 			}
431 
432 			sqset->eq_idx = idx;
433 
434 			idx = csio_wr_alloc_q(hw, CSIO_SCSI_IQSIZE,
435 					     CSIO_SCSI_IQ_WRSZ, CSIO_INGRESS,
436 					     (void *)hw, 0, 0,
437 					     csio_scsi_intx_handler);
438 			if (idx == -1) {
439 				csio_err(hw, "IQ creation failed for idx:%d\n",
440 					    idx);
441 				goto intr_disable;
442 			}
443 			sqset->iq_idx = idx;
444 		} /* for all CPUs */
445 	} /* For all ports */
446 
447 	hw->flags |= CSIO_HWF_Q_MEM_ALLOCED;
448 
449 	rv = csio_create_queues(hw);
450 	if (rv != 0)
451 		goto intr_disable;
452 
453 	/*
454 	 * Now request IRQs for the vectors. In the event of a failure,
455 	 * cleanup is handled internally by this function.
456 	 */
457 	rv = csio_request_irqs(hw);
458 	if (rv != 0)
459 		return -EINVAL;
460 
461 	return 0;
462 
463 intr_disable:
464 	csio_intr_disable(hw, false);
465 
466 	return -EINVAL;
467 }
468 
469 static int
470 csio_resource_alloc(struct csio_hw *hw)
471 {
472 	struct csio_wrm *wrm = csio_hw_to_wrm(hw);
473 	int rv = -ENOMEM;
474 
475 	wrm->num_q = ((CSIO_MAX_SCSI_QSETS * 2) + CSIO_HW_NIQ +
476 		       CSIO_HW_NEQ + CSIO_HW_NFLQ + CSIO_HW_NINTXQ);
477 
478 	hw->mb_mempool = mempool_create_kmalloc_pool(CSIO_MIN_MEMPOOL_SZ,
479 						  sizeof(struct csio_mb));
480 	if (!hw->mb_mempool)
481 		goto err;
482 
483 	hw->rnode_mempool = mempool_create_kmalloc_pool(CSIO_MIN_MEMPOOL_SZ,
484 						     sizeof(struct csio_rnode));
485 	if (!hw->rnode_mempool)
486 		goto err_free_mb_mempool;
487 
488 	hw->scsi_dma_pool = dma_pool_create("csio_scsi_dma_pool",
489 					    &hw->pdev->dev, CSIO_SCSI_RSP_LEN,
490 					    8, 0);
491 	if (!hw->scsi_dma_pool)
492 		goto err_free_rn_pool;
493 
494 	return 0;
495 
496 err_free_rn_pool:
497 	mempool_destroy(hw->rnode_mempool);
498 	hw->rnode_mempool = NULL;
499 err_free_mb_mempool:
500 	mempool_destroy(hw->mb_mempool);
501 	hw->mb_mempool = NULL;
502 err:
503 	return rv;
504 }
505 
506 static void
507 csio_resource_free(struct csio_hw *hw)
508 {
509 	dma_pool_destroy(hw->scsi_dma_pool);
510 	hw->scsi_dma_pool = NULL;
511 	mempool_destroy(hw->rnode_mempool);
512 	hw->rnode_mempool = NULL;
513 	mempool_destroy(hw->mb_mempool);
514 	hw->mb_mempool = NULL;
515 }
516 
517 /*
518  * csio_hw_alloc - Allocate and initialize the HW module.
519  * @pdev: PCI device.
520  *
521  * Allocates HW structure, DMA, memory resources, maps BARS to
522  * host memory and initializes HW module.
523  */
524 static struct csio_hw *csio_hw_alloc(struct pci_dev *pdev)
525 {
526 	struct csio_hw *hw;
527 
528 	hw = kzalloc(sizeof(struct csio_hw), GFP_KERNEL);
529 	if (!hw)
530 		goto err;
531 
532 	hw->pdev = pdev;
533 	strncpy(hw->drv_version, CSIO_DRV_VERSION, 32);
534 
535 	/* memory pool/DMA pool allocation */
536 	if (csio_resource_alloc(hw))
537 		goto err_free_hw;
538 
539 	/* Get the start address of registers from BAR 0 */
540 	hw->regstart = ioremap_nocache(pci_resource_start(pdev, 0),
541 				       pci_resource_len(pdev, 0));
542 	if (!hw->regstart) {
543 		csio_err(hw, "Could not map BAR 0, regstart = %p\n",
544 			 hw->regstart);
545 		goto err_resource_free;
546 	}
547 
548 	csio_hw_init_workers(hw);
549 
550 	if (csio_hw_init(hw))
551 		goto err_unmap_bar;
552 
553 	csio_dfs_create(hw);
554 
555 	csio_dbg(hw, "hw:%p\n", hw);
556 
557 	return hw;
558 
559 err_unmap_bar:
560 	csio_hw_exit_workers(hw);
561 	iounmap(hw->regstart);
562 err_resource_free:
563 	csio_resource_free(hw);
564 err_free_hw:
565 	kfree(hw);
566 err:
567 	return NULL;
568 }
569 
570 /*
571  * csio_hw_free - Uninitialize and free the HW module.
572  * @hw: The HW module
573  *
574  * Disable interrupts, uninit the HW module, free resources, free hw.
575  */
576 static void
577 csio_hw_free(struct csio_hw *hw)
578 {
579 	csio_intr_disable(hw, true);
580 	csio_hw_exit_workers(hw);
581 	csio_hw_exit(hw);
582 	iounmap(hw->regstart);
583 	csio_dfs_destroy(hw);
584 	csio_resource_free(hw);
585 	kfree(hw);
586 }
587 
588 /**
589  * csio_shost_init - Create and initialize the lnode module.
590  * @hw:		The HW module.
591  * @dev:	The device associated with this invocation.
592  * @probe:	Called from probe context or not?
593  * @os_pln:	Parent lnode if any.
594  *
595  * Allocates lnode structure via scsi_host_alloc, initializes
596  * shost, initializes lnode module and registers with SCSI ML
597  * via scsi_host_add. This function is shared between physical and
598  * virtual node ports.
599  */
600 struct csio_lnode *
601 csio_shost_init(struct csio_hw *hw, struct device *dev,
602 		  bool probe, struct csio_lnode *pln)
603 {
604 	struct Scsi_Host  *shost = NULL;
605 	struct csio_lnode *ln;
606 
607 	csio_fcoe_shost_template.cmd_per_lun = csio_lun_qdepth;
608 	csio_fcoe_shost_vport_template.cmd_per_lun = csio_lun_qdepth;
609 
610 	/*
611 	 * hw->pdev is the physical port's PCI dev structure,
612 	 * which will be different from the NPIV dev structure.
613 	 */
614 	if (dev == &hw->pdev->dev)
615 		shost = scsi_host_alloc(
616 				&csio_fcoe_shost_template,
617 				sizeof(struct csio_lnode));
618 	else
619 		shost = scsi_host_alloc(
620 				&csio_fcoe_shost_vport_template,
621 				sizeof(struct csio_lnode));
622 
623 	if (!shost)
624 		goto err;
625 
626 	ln = shost_priv(shost);
627 	memset(ln, 0, sizeof(struct csio_lnode));
628 
629 	/* Link common lnode to this lnode */
630 	ln->dev_num = (shost->host_no << 16);
631 
632 	shost->can_queue = CSIO_MAX_QUEUE;
633 	shost->this_id = -1;
634 	shost->unique_id = shost->host_no;
635 	shost->max_cmd_len = 16; /* Max CDB length supported */
636 	shost->max_id = min_t(uint32_t, csio_fcoe_rnodes,
637 			      hw->fres_info.max_ssns);
638 	shost->max_lun = CSIO_MAX_LUN;
639 	if (dev == &hw->pdev->dev)
640 		shost->transportt = csio_fcoe_transport;
641 	else
642 		shost->transportt = csio_fcoe_transport_vport;
643 
644 	/* root lnode */
645 	if (!hw->rln)
646 		hw->rln = ln;
647 
648 	/* Other initialization here: Common, Transport specific */
649 	if (csio_lnode_init(ln, hw, pln))
650 		goto err_shost_put;
651 
652 	if (scsi_add_host(shost, dev))
653 		goto err_lnode_exit;
654 
655 	return ln;
656 
657 err_lnode_exit:
658 	csio_lnode_exit(ln);
659 err_shost_put:
660 	scsi_host_put(shost);
661 err:
662 	return NULL;
663 }
664 
665 /**
666  * csio_shost_exit - De-instantiate the shost.
667  * @ln:		The lnode module corresponding to the shost.
668  *
669  */
670 void
671 csio_shost_exit(struct csio_lnode *ln)
672 {
673 	struct Scsi_Host *shost = csio_ln_to_shost(ln);
674 	struct csio_hw *hw = csio_lnode_to_hw(ln);
675 
676 	/* Inform transport */
677 	fc_remove_host(shost);
678 
679 	/* Inform SCSI ML */
680 	scsi_remove_host(shost);
681 
682 	/* Flush all the events, so that any rnode removal events
683 	 * already queued are all handled, before we remove the lnode.
684 	 */
685 	spin_lock_irq(&hw->lock);
686 	csio_evtq_flush(hw);
687 	spin_unlock_irq(&hw->lock);
688 
689 	csio_lnode_exit(ln);
690 	scsi_host_put(shost);
691 }
692 
693 struct csio_lnode *
694 csio_lnode_alloc(struct csio_hw *hw)
695 {
696 	return csio_shost_init(hw, &hw->pdev->dev, false, NULL);
697 }
698 
699 void
700 csio_lnodes_block_request(struct csio_hw *hw)
701 {
702 	struct Scsi_Host  *shost;
703 	struct csio_lnode *sln;
704 	struct csio_lnode *ln;
705 	struct list_head *cur_ln, *cur_cln;
706 	struct csio_lnode **lnode_list;
707 	int cur_cnt = 0, ii;
708 
709 	lnode_list = kzalloc((sizeof(struct csio_lnode *) * hw->num_lns),
710 			GFP_KERNEL);
711 	if (!lnode_list) {
712 		csio_err(hw, "Failed to allocate lnodes_list");
713 		return;
714 	}
715 
716 	spin_lock_irq(&hw->lock);
717 	/* Traverse sibling lnodes */
718 	list_for_each(cur_ln, &hw->sln_head) {
719 		sln = (struct csio_lnode *) cur_ln;
720 		lnode_list[cur_cnt++] = sln;
721 
722 		/* Traverse children lnodes */
723 		list_for_each(cur_cln, &sln->cln_head)
724 			lnode_list[cur_cnt++] = (struct csio_lnode *) cur_cln;
725 	}
726 	spin_unlock_irq(&hw->lock);
727 
728 	for (ii = 0; ii < cur_cnt; ii++) {
729 		csio_dbg(hw, "Blocking IOs on lnode: %p\n", lnode_list[ii]);
730 		ln = lnode_list[ii];
731 		shost = csio_ln_to_shost(ln);
732 		scsi_block_requests(shost);
733 
734 	}
735 	kfree(lnode_list);
736 }
737 
738 void
739 csio_lnodes_unblock_request(struct csio_hw *hw)
740 {
741 	struct csio_lnode *ln;
742 	struct Scsi_Host  *shost;
743 	struct csio_lnode *sln;
744 	struct list_head *cur_ln, *cur_cln;
745 	struct csio_lnode **lnode_list;
746 	int cur_cnt = 0, ii;
747 
748 	lnode_list = kzalloc((sizeof(struct csio_lnode *) * hw->num_lns),
749 			GFP_KERNEL);
750 	if (!lnode_list) {
751 		csio_err(hw, "Failed to allocate lnodes_list");
752 		return;
753 	}
754 
755 	spin_lock_irq(&hw->lock);
756 	/* Traverse sibling lnodes */
757 	list_for_each(cur_ln, &hw->sln_head) {
758 		sln = (struct csio_lnode *) cur_ln;
759 		lnode_list[cur_cnt++] = sln;
760 
761 		/* Traverse children lnodes */
762 		list_for_each(cur_cln, &sln->cln_head)
763 			lnode_list[cur_cnt++] = (struct csio_lnode *) cur_cln;
764 	}
765 	spin_unlock_irq(&hw->lock);
766 
767 	for (ii = 0; ii < cur_cnt; ii++) {
768 		csio_dbg(hw, "unblocking IOs on lnode: %p\n", lnode_list[ii]);
769 		ln = lnode_list[ii];
770 		shost = csio_ln_to_shost(ln);
771 		scsi_unblock_requests(shost);
772 	}
773 	kfree(lnode_list);
774 }
775 
776 void
777 csio_lnodes_block_by_port(struct csio_hw *hw, uint8_t portid)
778 {
779 	struct csio_lnode *ln;
780 	struct Scsi_Host  *shost;
781 	struct csio_lnode *sln;
782 	struct list_head *cur_ln, *cur_cln;
783 	struct csio_lnode **lnode_list;
784 	int cur_cnt = 0, ii;
785 
786 	lnode_list = kzalloc((sizeof(struct csio_lnode *) * hw->num_lns),
787 			GFP_KERNEL);
788 	if (!lnode_list) {
789 		csio_err(hw, "Failed to allocate lnodes_list");
790 		return;
791 	}
792 
793 	spin_lock_irq(&hw->lock);
794 	/* Traverse sibling lnodes */
795 	list_for_each(cur_ln, &hw->sln_head) {
796 		sln = (struct csio_lnode *) cur_ln;
797 		if (sln->portid != portid)
798 			continue;
799 
800 		lnode_list[cur_cnt++] = sln;
801 
802 		/* Traverse children lnodes */
803 		list_for_each(cur_cln, &sln->cln_head)
804 			lnode_list[cur_cnt++] = (struct csio_lnode *) cur_cln;
805 	}
806 	spin_unlock_irq(&hw->lock);
807 
808 	for (ii = 0; ii < cur_cnt; ii++) {
809 		csio_dbg(hw, "Blocking IOs on lnode: %p\n", lnode_list[ii]);
810 		ln = lnode_list[ii];
811 		shost = csio_ln_to_shost(ln);
812 		scsi_block_requests(shost);
813 	}
814 	kfree(lnode_list);
815 }
816 
817 void
818 csio_lnodes_unblock_by_port(struct csio_hw *hw, uint8_t portid)
819 {
820 	struct csio_lnode *ln;
821 	struct Scsi_Host  *shost;
822 	struct csio_lnode *sln;
823 	struct list_head *cur_ln, *cur_cln;
824 	struct csio_lnode **lnode_list;
825 	int cur_cnt = 0, ii;
826 
827 	lnode_list = kzalloc((sizeof(struct csio_lnode *) * hw->num_lns),
828 			GFP_KERNEL);
829 	if (!lnode_list) {
830 		csio_err(hw, "Failed to allocate lnodes_list");
831 		return;
832 	}
833 
834 	spin_lock_irq(&hw->lock);
835 	/* Traverse sibling lnodes */
836 	list_for_each(cur_ln, &hw->sln_head) {
837 		sln = (struct csio_lnode *) cur_ln;
838 		if (sln->portid != portid)
839 			continue;
840 		lnode_list[cur_cnt++] = sln;
841 
842 		/* Traverse children lnodes */
843 		list_for_each(cur_cln, &sln->cln_head)
844 			lnode_list[cur_cnt++] = (struct csio_lnode *) cur_cln;
845 	}
846 	spin_unlock_irq(&hw->lock);
847 
848 	for (ii = 0; ii < cur_cnt; ii++) {
849 		csio_dbg(hw, "unblocking IOs on lnode: %p\n", lnode_list[ii]);
850 		ln = lnode_list[ii];
851 		shost = csio_ln_to_shost(ln);
852 		scsi_unblock_requests(shost);
853 	}
854 	kfree(lnode_list);
855 }
856 
857 void
858 csio_lnodes_exit(struct csio_hw *hw, bool npiv)
859 {
860 	struct csio_lnode *sln;
861 	struct csio_lnode *ln;
862 	struct list_head *cur_ln, *cur_cln;
863 	struct csio_lnode **lnode_list;
864 	int cur_cnt = 0, ii;
865 
866 	lnode_list = kzalloc((sizeof(struct csio_lnode *) * hw->num_lns),
867 			GFP_KERNEL);
868 	if (!lnode_list) {
869 		csio_err(hw, "lnodes_exit: Failed to allocate lnodes_list.\n");
870 		return;
871 	}
872 
873 	/* Get all child lnodes(NPIV ports) */
874 	spin_lock_irq(&hw->lock);
875 	list_for_each(cur_ln, &hw->sln_head) {
876 		sln = (struct csio_lnode *) cur_ln;
877 
878 		/* Traverse children lnodes */
879 		list_for_each(cur_cln, &sln->cln_head)
880 			lnode_list[cur_cnt++] = (struct csio_lnode *) cur_cln;
881 	}
882 	spin_unlock_irq(&hw->lock);
883 
884 	/* Delete NPIV lnodes */
885 	for (ii = 0; ii < cur_cnt; ii++) {
886 		csio_dbg(hw, "Deleting child lnode: %p\n", lnode_list[ii]);
887 		ln = lnode_list[ii];
888 		fc_vport_terminate(ln->fc_vport);
889 	}
890 
891 	/* Delete only npiv lnodes */
892 	if (npiv)
893 		goto free_lnodes;
894 
895 	cur_cnt = 0;
896 	/* Get all physical lnodes */
897 	spin_lock_irq(&hw->lock);
898 	/* Traverse sibling lnodes */
899 	list_for_each(cur_ln, &hw->sln_head) {
900 		sln = (struct csio_lnode *) cur_ln;
901 		lnode_list[cur_cnt++] = sln;
902 	}
903 	spin_unlock_irq(&hw->lock);
904 
905 	/* Delete physical lnodes */
906 	for (ii = 0; ii < cur_cnt; ii++) {
907 		csio_dbg(hw, "Deleting parent lnode: %p\n", lnode_list[ii]);
908 		csio_shost_exit(lnode_list[ii]);
909 	}
910 
911 free_lnodes:
912 	kfree(lnode_list);
913 }
914 
915 /*
916  * csio_lnode_init_post: Set lnode attributes after starting HW.
917  * @ln: lnode.
918  *
919  */
920 static void
921 csio_lnode_init_post(struct csio_lnode *ln)
922 {
923 	struct Scsi_Host  *shost = csio_ln_to_shost(ln);
924 
925 	csio_fchost_attr_init(ln);
926 
927 	scsi_scan_host(shost);
928 }
929 
930 /*
931  * csio_probe_one - Instantiate this function.
932  * @pdev: PCI device
933  * @id: Device ID
934  *
935  * This is the .probe() callback of the driver. This function:
936  * - Initializes the PCI function by enabling MMIO, setting bus
937  *   mastership and setting DMA mask.
938  * - Allocates HW structure, DMA, memory resources, maps BARS to
939  *   host memory and initializes HW module.
940  * - Allocates lnode structure via scsi_host_alloc, initializes
941  *   shost, initialized lnode module and registers with SCSI ML
942  *   via scsi_host_add.
943  * - Enables interrupts, and starts the chip by kicking off the
944  *   HW state machine.
945  * - Once hardware is ready, initiated scan of the host via
946  *   scsi_scan_host.
947  */
948 static int csio_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
949 {
950 	int rv;
951 	int bars;
952 	int i;
953 	struct csio_hw *hw;
954 	struct csio_lnode *ln;
955 
956 	/* probe only T5 and T6 cards */
957 	if (!csio_is_t5((pdev->device & CSIO_HW_CHIP_MASK)) &&
958 	    !csio_is_t6((pdev->device & CSIO_HW_CHIP_MASK)))
959 		return -ENODEV;
960 
961 	rv = csio_pci_init(pdev, &bars);
962 	if (rv)
963 		goto err;
964 
965 	hw = csio_hw_alloc(pdev);
966 	if (!hw) {
967 		rv = -ENODEV;
968 		goto err_pci_exit;
969 	}
970 
971 	if (!pcie_relaxed_ordering_enabled(pdev))
972 		hw->flags |= CSIO_HWF_ROOT_NO_RELAXED_ORDERING;
973 
974 	pci_set_drvdata(pdev, hw);
975 
976 	rv = csio_hw_start(hw);
977 	if (rv) {
978 		if (rv == -EINVAL) {
979 			dev_err(&pdev->dev,
980 				"Failed to start FW, continuing in debug mode.\n");
981 			return 0;
982 		}
983 		goto err_lnode_exit;
984 	}
985 
986 	sprintf(hw->fwrev_str, "%u.%u.%u.%u\n",
987 		    FW_HDR_FW_VER_MAJOR_G(hw->fwrev),
988 		    FW_HDR_FW_VER_MINOR_G(hw->fwrev),
989 		    FW_HDR_FW_VER_MICRO_G(hw->fwrev),
990 		    FW_HDR_FW_VER_BUILD_G(hw->fwrev));
991 
992 	for (i = 0; i < hw->num_pports; i++) {
993 		ln = csio_shost_init(hw, &pdev->dev, true, NULL);
994 		if (!ln) {
995 			rv = -ENODEV;
996 			break;
997 		}
998 		/* Initialize portid */
999 		ln->portid = hw->pport[i].portid;
1000 
1001 		spin_lock_irq(&hw->lock);
1002 		if (csio_lnode_start(ln) != 0)
1003 			rv = -ENODEV;
1004 		spin_unlock_irq(&hw->lock);
1005 
1006 		if (rv)
1007 			break;
1008 
1009 		csio_lnode_init_post(ln);
1010 	}
1011 
1012 	if (rv)
1013 		goto err_lnode_exit;
1014 
1015 	return 0;
1016 
1017 err_lnode_exit:
1018 	csio_lnodes_block_request(hw);
1019 	spin_lock_irq(&hw->lock);
1020 	csio_hw_stop(hw);
1021 	spin_unlock_irq(&hw->lock);
1022 	csio_lnodes_unblock_request(hw);
1023 	csio_lnodes_exit(hw, 0);
1024 	csio_hw_free(hw);
1025 err_pci_exit:
1026 	csio_pci_exit(pdev, &bars);
1027 err:
1028 	dev_err(&pdev->dev, "probe of device failed: %d\n", rv);
1029 	return rv;
1030 }
1031 
1032 /*
1033  * csio_remove_one - Remove one instance of the driver at this PCI function.
1034  * @pdev: PCI device
1035  *
1036  * Used during hotplug operation.
1037  */
1038 static void csio_remove_one(struct pci_dev *pdev)
1039 {
1040 	struct csio_hw *hw = pci_get_drvdata(pdev);
1041 	int bars = pci_select_bars(pdev, IORESOURCE_MEM);
1042 
1043 	csio_lnodes_block_request(hw);
1044 	spin_lock_irq(&hw->lock);
1045 
1046 	/* Stops lnode, Rnode s/m
1047 	 * Quiesce IOs.
1048 	 * All sessions with remote ports are unregistered.
1049 	 */
1050 	csio_hw_stop(hw);
1051 	spin_unlock_irq(&hw->lock);
1052 	csio_lnodes_unblock_request(hw);
1053 
1054 	csio_lnodes_exit(hw, 0);
1055 	csio_hw_free(hw);
1056 	csio_pci_exit(pdev, &bars);
1057 }
1058 
1059 /*
1060  * csio_pci_error_detected - PCI error was detected
1061  * @pdev: PCI device
1062  *
1063  */
1064 static pci_ers_result_t
1065 csio_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
1066 {
1067 	struct csio_hw *hw = pci_get_drvdata(pdev);
1068 
1069 	csio_lnodes_block_request(hw);
1070 	spin_lock_irq(&hw->lock);
1071 
1072 	/* Post PCI error detected evt to HW s/m
1073 	 * HW s/m handles this evt by quiescing IOs, unregisters rports
1074 	 * and finally takes the device to offline.
1075 	 */
1076 	csio_post_event(&hw->sm, CSIO_HWE_PCIERR_DETECTED);
1077 	spin_unlock_irq(&hw->lock);
1078 	csio_lnodes_unblock_request(hw);
1079 	csio_lnodes_exit(hw, 0);
1080 	csio_intr_disable(hw, true);
1081 	pci_disable_device(pdev);
1082 	return state == pci_channel_io_perm_failure ?
1083 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
1084 }
1085 
1086 /*
1087  * csio_pci_slot_reset - PCI slot has been reset.
1088  * @pdev: PCI device
1089  *
1090  */
1091 static pci_ers_result_t
1092 csio_pci_slot_reset(struct pci_dev *pdev)
1093 {
1094 	struct csio_hw *hw = pci_get_drvdata(pdev);
1095 	int ready;
1096 
1097 	if (pci_enable_device(pdev)) {
1098 		dev_err(&pdev->dev, "cannot re-enable device in slot reset\n");
1099 		return PCI_ERS_RESULT_DISCONNECT;
1100 	}
1101 
1102 	pci_set_master(pdev);
1103 	pci_restore_state(pdev);
1104 	pci_save_state(pdev);
1105 	pci_cleanup_aer_uncorrect_error_status(pdev);
1106 
1107 	/* Bring HW s/m to ready state.
1108 	 * but don't resume IOs.
1109 	 */
1110 	spin_lock_irq(&hw->lock);
1111 	csio_post_event(&hw->sm, CSIO_HWE_PCIERR_SLOT_RESET);
1112 	ready = csio_is_hw_ready(hw);
1113 	spin_unlock_irq(&hw->lock);
1114 
1115 	if (ready) {
1116 		return PCI_ERS_RESULT_RECOVERED;
1117 	} else {
1118 		dev_err(&pdev->dev, "Can't initialize HW when in slot reset\n");
1119 		return PCI_ERS_RESULT_DISCONNECT;
1120 	}
1121 }
1122 
1123 /*
1124  * csio_pci_resume - Resume normal operations
1125  * @pdev: PCI device
1126  *
1127  */
1128 static void
1129 csio_pci_resume(struct pci_dev *pdev)
1130 {
1131 	struct csio_hw *hw = pci_get_drvdata(pdev);
1132 	struct csio_lnode *ln;
1133 	int rv = 0;
1134 	int i;
1135 
1136 	/* Bring the LINK UP and Resume IO */
1137 
1138 	for (i = 0; i < hw->num_pports; i++) {
1139 		ln = csio_shost_init(hw, &pdev->dev, true, NULL);
1140 		if (!ln) {
1141 			rv = -ENODEV;
1142 			break;
1143 		}
1144 		/* Initialize portid */
1145 		ln->portid = hw->pport[i].portid;
1146 
1147 		spin_lock_irq(&hw->lock);
1148 		if (csio_lnode_start(ln) != 0)
1149 			rv = -ENODEV;
1150 		spin_unlock_irq(&hw->lock);
1151 
1152 		if (rv)
1153 			break;
1154 
1155 		csio_lnode_init_post(ln);
1156 	}
1157 
1158 	if (rv)
1159 		goto err_resume_exit;
1160 
1161 	return;
1162 
1163 err_resume_exit:
1164 	csio_lnodes_block_request(hw);
1165 	spin_lock_irq(&hw->lock);
1166 	csio_hw_stop(hw);
1167 	spin_unlock_irq(&hw->lock);
1168 	csio_lnodes_unblock_request(hw);
1169 	csio_lnodes_exit(hw, 0);
1170 	csio_hw_free(hw);
1171 	dev_err(&pdev->dev, "resume of device failed: %d\n", rv);
1172 }
1173 
1174 static struct pci_error_handlers csio_err_handler = {
1175 	.error_detected = csio_pci_error_detected,
1176 	.slot_reset	= csio_pci_slot_reset,
1177 	.resume		= csio_pci_resume,
1178 };
1179 
1180 /*
1181  *  Macros needed to support the PCI Device ID Table ...
1182  */
1183 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
1184 	static const struct pci_device_id csio_pci_tbl[] = {
1185 /* Define for FCoE uses PF6 */
1186 #define CH_PCI_DEVICE_ID_FUNCTION	0x6
1187 
1188 #define CH_PCI_ID_TABLE_ENTRY(devid) \
1189 		{ PCI_VDEVICE(CHELSIO, (devid)), 0 }
1190 
1191 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END { 0, } }
1192 
1193 #include "t4_pci_id_tbl.h"
1194 
1195 static struct pci_driver csio_pci_driver = {
1196 	.name		= KBUILD_MODNAME,
1197 	.driver		= {
1198 		.owner	= THIS_MODULE,
1199 	},
1200 	.id_table	= csio_pci_tbl,
1201 	.probe		= csio_probe_one,
1202 	.remove		= csio_remove_one,
1203 	.err_handler	= &csio_err_handler,
1204 };
1205 
1206 /*
1207  * csio_init - Chelsio storage driver initialization function.
1208  *
1209  */
1210 static int __init
1211 csio_init(void)
1212 {
1213 	int rv = -ENOMEM;
1214 
1215 	pr_info("%s %s\n", CSIO_DRV_DESC, CSIO_DRV_VERSION);
1216 
1217 	csio_dfs_init();
1218 
1219 	csio_fcoe_transport = fc_attach_transport(&csio_fc_transport_funcs);
1220 	if (!csio_fcoe_transport)
1221 		goto err;
1222 
1223 	csio_fcoe_transport_vport =
1224 			fc_attach_transport(&csio_fc_transport_vport_funcs);
1225 	if (!csio_fcoe_transport_vport)
1226 		goto err_vport;
1227 
1228 	rv = pci_register_driver(&csio_pci_driver);
1229 	if (rv)
1230 		goto err_pci;
1231 
1232 	return 0;
1233 
1234 err_pci:
1235 	fc_release_transport(csio_fcoe_transport_vport);
1236 err_vport:
1237 	fc_release_transport(csio_fcoe_transport);
1238 err:
1239 	csio_dfs_exit();
1240 	return rv;
1241 }
1242 
1243 /*
1244  * csio_exit - Chelsio storage driver uninitialization .
1245  *
1246  * Function that gets called in the unload path.
1247  */
1248 static void __exit
1249 csio_exit(void)
1250 {
1251 	pci_unregister_driver(&csio_pci_driver);
1252 	csio_dfs_exit();
1253 	fc_release_transport(csio_fcoe_transport_vport);
1254 	fc_release_transport(csio_fcoe_transport);
1255 }
1256 
1257 module_init(csio_init);
1258 module_exit(csio_exit);
1259 MODULE_AUTHOR(CSIO_DRV_AUTHOR);
1260 MODULE_DESCRIPTION(CSIO_DRV_DESC);
1261 MODULE_LICENSE(CSIO_DRV_LICENSE);
1262 MODULE_DEVICE_TABLE(pci, csio_pci_tbl);
1263 MODULE_VERSION(CSIO_DRV_VERSION);
1264 MODULE_FIRMWARE(FW_FNAME_T5);
1265 MODULE_FIRMWARE(FW_FNAME_T6);
1266