1 /**********************************************************************
2  * Author: Cavium, Inc.
3  *
4  * Contact: support@cavium.com
5  *          Please include "LiquidIO" in the subject.
6  *
7  * Copyright (c) 2003-2016 Cavium, Inc.
8  *
9  * This file is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License, Version 2, as
11  * published by the Free Software Foundation.
12  *
13  * This file is distributed in the hope that it will be useful, but
14  * AS-IS and WITHOUT ANY WARRANTY; without even the implied warranty
15  * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE, TITLE, or
16  * NONINFRINGEMENT.  See the GNU General Public License for more
17  * details.
18  **********************************************************************/
19 #include <linux/pci.h>
20 #include <linux/netdevice.h>
21 #include "liquidio_common.h"
22 #include "octeon_droq.h"
23 #include "octeon_iq.h"
24 #include "response_manager.h"
25 #include "octeon_device.h"
26 #include "octeon_main.h"
27 
28 static void oct_poll_req_completion(struct work_struct *work);
29 
30 int octeon_setup_response_list(struct octeon_device *oct)
31 {
32 	int i, ret = 0;
33 	struct cavium_wq *cwq;
34 
35 	for (i = 0; i < MAX_RESPONSE_LISTS; i++) {
36 		INIT_LIST_HEAD(&oct->response_list[i].head);
37 		spin_lock_init(&oct->response_list[i].lock);
38 		atomic_set(&oct->response_list[i].pending_req_count, 0);
39 	}
40 	spin_lock_init(&oct->cmd_resp_wqlock);
41 
42 	oct->dma_comp_wq.wq = alloc_workqueue("dma-comp", WQ_MEM_RECLAIM, 0);
43 	if (!oct->dma_comp_wq.wq) {
44 		dev_err(&oct->pci_dev->dev, "failed to create wq thread\n");
45 		return -ENOMEM;
46 	}
47 
48 	cwq = &oct->dma_comp_wq;
49 	INIT_DELAYED_WORK(&cwq->wk.work, oct_poll_req_completion);
50 	cwq->wk.ctxptr = oct;
51 	oct->cmd_resp_state = OCT_DRV_ONLINE;
52 
53 	return ret;
54 }
55 EXPORT_SYMBOL_GPL(octeon_setup_response_list);
56 
57 void octeon_delete_response_list(struct octeon_device *oct)
58 {
59 	cancel_delayed_work_sync(&oct->dma_comp_wq.wk.work);
60 	destroy_workqueue(oct->dma_comp_wq.wq);
61 }
62 EXPORT_SYMBOL_GPL(octeon_delete_response_list);
63 
64 int lio_process_ordered_list(struct octeon_device *octeon_dev,
65 			     u32 force_quit)
66 {
67 	struct octeon_response_list *ordered_sc_list;
68 	struct octeon_soft_command *sc;
69 	int request_complete = 0;
70 	int resp_to_process = MAX_ORD_REQS_TO_PROCESS;
71 	u32 status;
72 	u64 status64;
73 
74 	octeon_free_sc_done_list(octeon_dev);
75 
76 	ordered_sc_list = &octeon_dev->response_list[OCTEON_ORDERED_SC_LIST];
77 
78 	do {
79 		spin_lock_bh(&ordered_sc_list->lock);
80 
81 		if (list_empty(&ordered_sc_list->head)) {
82 			spin_unlock_bh(&ordered_sc_list->lock);
83 			return 1;
84 		}
85 
86 		sc = list_first_entry(&ordered_sc_list->head,
87 				      struct octeon_soft_command, node);
88 
89 		status = OCTEON_REQUEST_PENDING;
90 
91 		/* check if octeon has finished DMA'ing a response
92 		 * to where rptr is pointing to
93 		 */
94 		status64 = *sc->status_word;
95 
96 		if (status64 != COMPLETION_WORD_INIT) {
97 			/* This logic ensures that all 64b have been written.
98 			 * 1. check byte 0 for non-FF
99 			 * 2. if non-FF, then swap result from BE to host order
100 			 * 3. check byte 7 (swapped to 0) for non-FF
101 			 * 4. if non-FF, use the low 32-bit status code
102 			 * 5. if either byte 0 or byte 7 is FF, don't use status
103 			 */
104 			if ((status64 & 0xff) != 0xff) {
105 				octeon_swap_8B_data(&status64, 1);
106 				if (((status64 & 0xff) != 0xff)) {
107 					/* retrieve 16-bit firmware status */
108 					status = (u32)(status64 & 0xffffULL);
109 					if (status) {
110 						status =
111 						  FIRMWARE_STATUS_CODE(status);
112 					} else {
113 						/* i.e. no error */
114 						status = OCTEON_REQUEST_DONE;
115 					}
116 				}
117 			}
118 		} else if (unlikely(force_quit) || (sc->expiry_time &&
119 			time_after(jiffies, (unsigned long)sc->expiry_time))) {
120 			struct octeon_instr_irh *irh =
121 				(struct octeon_instr_irh *)&sc->cmd.cmd3.irh;
122 
123 			dev_err(&octeon_dev->pci_dev->dev, "%s: ", __func__);
124 			dev_err(&octeon_dev->pci_dev->dev,
125 				"cmd %x/%x/%llx/%llx failed, ",
126 				irh->opcode, irh->subcode,
127 				sc->cmd.cmd3.ossp[0], sc->cmd.cmd3.ossp[1]);
128 			dev_err(&octeon_dev->pci_dev->dev,
129 				"timeout (%ld, %ld)\n",
130 				(long)jiffies, (long)sc->expiry_time);
131 			status = OCTEON_REQUEST_TIMEOUT;
132 		}
133 
134 		if (status != OCTEON_REQUEST_PENDING) {
135 			sc->sc_status = status;
136 
137 			/* we have received a response or we have timed out */
138 			/* remove node from linked list */
139 			list_del(&sc->node);
140 			atomic_dec(&octeon_dev->response_list
141 				   [OCTEON_ORDERED_SC_LIST].
142 				   pending_req_count);
143 
144 			if (!sc->callback) {
145 				atomic_inc(&octeon_dev->response_list
146 					   [OCTEON_DONE_SC_LIST].
147 					   pending_req_count);
148 				list_add_tail(&sc->node,
149 					      &octeon_dev->response_list
150 					      [OCTEON_DONE_SC_LIST].head);
151 
152 				if (unlikely(READ_ONCE(sc->caller_is_done))) {
153 					/* caller does not wait for response
154 					 * from firmware
155 					 */
156 					if (status != OCTEON_REQUEST_DONE) {
157 						struct octeon_instr_irh *irh;
158 
159 						irh =
160 						    (struct octeon_instr_irh *)
161 						    &sc->cmd.cmd3.irh;
162 						dev_dbg
163 						    (&octeon_dev->pci_dev->dev,
164 						    "%s: sc failed: opcode=%x, ",
165 						    __func__, irh->opcode);
166 						dev_dbg
167 						    (&octeon_dev->pci_dev->dev,
168 						    "subcode=%x, ossp[0]=%llx, ",
169 						    irh->subcode,
170 						    sc->cmd.cmd3.ossp[0]);
171 						dev_dbg
172 						    (&octeon_dev->pci_dev->dev,
173 						    "ossp[1]=%llx, status=%d\n",
174 						    sc->cmd.cmd3.ossp[1],
175 						    status);
176 					}
177 				} else {
178 					complete(&sc->complete);
179 				}
180 
181 				spin_unlock_bh(&ordered_sc_list->lock);
182 			} else {
183 				/* sc with callback function */
184 				if (status == OCTEON_REQUEST_TIMEOUT) {
185 					atomic_inc(&octeon_dev->response_list
186 						   [OCTEON_ZOMBIE_SC_LIST].
187 						   pending_req_count);
188 					list_add_tail(&sc->node,
189 						      &octeon_dev->response_list
190 						      [OCTEON_ZOMBIE_SC_LIST].
191 						      head);
192 				}
193 
194 				spin_unlock_bh(&ordered_sc_list->lock);
195 
196 				sc->callback(octeon_dev, status,
197 					     sc->callback_arg);
198 				/* sc is freed by caller */
199 			}
200 
201 			request_complete++;
202 
203 		} else {
204 			/* no response yet */
205 			request_complete = 0;
206 			spin_unlock_bh
207 			    (&ordered_sc_list->lock);
208 		}
209 
210 		/* If we hit the Max Ordered requests to process every loop,
211 		 * we quit
212 		 * and let this function be invoked the next time the poll
213 		 * thread runs
214 		 * to process the remaining requests. This function can take up
215 		 * the entire CPU if there is no upper limit to the requests
216 		 * processed.
217 		 */
218 		if (request_complete >= resp_to_process)
219 			break;
220 	} while (request_complete);
221 
222 	return 0;
223 }
224 EXPORT_SYMBOL_GPL(lio_process_ordered_list);
225 
226 static void oct_poll_req_completion(struct work_struct *work)
227 {
228 	struct cavium_wk *wk = (struct cavium_wk *)work;
229 	struct octeon_device *oct = (struct octeon_device *)wk->ctxptr;
230 	struct cavium_wq *cwq = &oct->dma_comp_wq;
231 
232 	lio_process_ordered_list(oct, 0);
233 
234 	if (atomic_read(&oct->response_list
235 			[OCTEON_ORDERED_SC_LIST].pending_req_count))
236 		queue_delayed_work(cwq->wq, &cwq->wk.work, msecs_to_jiffies(1));
237 }
238