Commit | Line | Data |
---|---|---|
f21fb3ed RV |
1 | /********************************************************************** |
2 | * Author: Cavium, Inc. | |
3 | * | |
4 | * Contact: support@cavium.com | |
5 | * Please include "LiquidIO" in the subject. | |
6 | * | |
7 | * Copyright (c) 2003-2015 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 | * This file may also be available under a different license from Cavium. | |
20 | * Contact Cavium, Inc. for more information | |
21 | **********************************************************************/ | |
22 | #include <linux/version.h> | |
23 | #include <linux/types.h> | |
24 | #include <linux/list.h> | |
25 | #include <linux/interrupt.h> | |
26 | #include <linux/pci.h> | |
27 | #include <linux/kthread.h> | |
28 | #include <linux/netdevice.h> | |
5b173cf9 | 29 | #include <linux/vmalloc.h> |
f21fb3ed RV |
30 | #include "octeon_config.h" |
31 | #include "liquidio_common.h" | |
32 | #include "octeon_droq.h" | |
33 | #include "octeon_iq.h" | |
34 | #include "response_manager.h" | |
35 | #include "octeon_device.h" | |
36 | #include "octeon_nic.h" | |
37 | #include "octeon_main.h" | |
38 | #include "octeon_network.h" | |
39 | #include "cn66xx_regs.h" | |
40 | #include "cn66xx_device.h" | |
41 | #include "cn68xx_regs.h" | |
42 | #include "cn68xx_device.h" | |
43 | #include "liquidio_image.h" | |
44 | ||
45 | #define INCR_INSTRQUEUE_PKT_COUNT(octeon_dev_ptr, iq_no, field, count) \ | |
46 | (octeon_dev_ptr->instr_queue[iq_no]->stats.field += count) | |
47 | ||
48 | struct iq_post_status { | |
49 | int status; | |
50 | int index; | |
51 | }; | |
52 | ||
53 | static void check_db_timeout(struct work_struct *work); | |
54 | static void __check_db_timeout(struct octeon_device *oct, unsigned long iq_no); | |
55 | ||
56 | static void (*reqtype_free_fn[MAX_OCTEON_DEVICES][REQTYPE_LAST + 1]) (void *); | |
57 | ||
58 | static inline int IQ_INSTR_MODE_64B(struct octeon_device *oct, int iq_no) | |
59 | { | |
60 | struct octeon_instr_queue *iq = | |
61 | (struct octeon_instr_queue *)oct->instr_queue[iq_no]; | |
62 | return iq->iqcmd_64B; | |
63 | } | |
64 | ||
65 | #define IQ_INSTR_MODE_32B(oct, iq_no) (!IQ_INSTR_MODE_64B(oct, iq_no)) | |
66 | ||
67 | /* Define this to return the request status comaptible to old code */ | |
68 | /*#define OCTEON_USE_OLD_REQ_STATUS*/ | |
69 | ||
70 | /* Return 0 on success, 1 on failure */ | |
71 | int octeon_init_instr_queue(struct octeon_device *oct, | |
26236fa9 RV |
72 | union oct_txpciq txpciq, |
73 | u32 num_descs) | |
f21fb3ed RV |
74 | { |
75 | struct octeon_instr_queue *iq; | |
76 | struct octeon_iq_config *conf = NULL; | |
26236fa9 | 77 | u32 iq_no = (u32)txpciq.s.q_no; |
f21fb3ed RV |
78 | u32 q_size; |
79 | struct cavium_wq *db_wq; | |
26236fa9 RV |
80 | int orig_node = dev_to_node(&oct->pci_dev->dev); |
81 | int numa_node = cpu_to_node(iq_no % num_online_cpus()); | |
f21fb3ed RV |
82 | |
83 | if (OCTEON_CN6XXX(oct)) | |
84 | conf = &(CFG_GET_IQ_CFG(CHIP_FIELD(oct, cn6xxx, conf))); | |
85 | ||
86 | if (!conf) { | |
87 | dev_err(&oct->pci_dev->dev, "Unsupported Chip %x\n", | |
88 | oct->chip_id); | |
89 | return 1; | |
90 | } | |
91 | ||
92 | if (num_descs & (num_descs - 1)) { | |
93 | dev_err(&oct->pci_dev->dev, | |
94 | "Number of descriptors for instr queue %d not in power of 2.\n", | |
95 | iq_no); | |
96 | return 1; | |
97 | } | |
98 | ||
99 | q_size = (u32)conf->instr_type * num_descs; | |
100 | ||
101 | iq = oct->instr_queue[iq_no]; | |
6a885b60 | 102 | iq->oct_dev = oct; |
f21fb3ed | 103 | |
26236fa9 | 104 | set_dev_node(&oct->pci_dev->dev, numa_node); |
f21fb3ed RV |
105 | iq->base_addr = lio_dma_alloc(oct, q_size, |
106 | (dma_addr_t *)&iq->base_addr_dma); | |
26236fa9 RV |
107 | set_dev_node(&oct->pci_dev->dev, orig_node); |
108 | if (!iq->base_addr) | |
109 | iq->base_addr = lio_dma_alloc(oct, q_size, | |
110 | (dma_addr_t *)&iq->base_addr_dma); | |
f21fb3ed RV |
111 | if (!iq->base_addr) { |
112 | dev_err(&oct->pci_dev->dev, "Cannot allocate memory for instr queue %d\n", | |
113 | iq_no); | |
114 | return 1; | |
115 | } | |
116 | ||
117 | iq->max_count = num_descs; | |
118 | ||
119 | /* Initialize a list to holds requests that have been posted to Octeon | |
120 | * but has yet to be fetched by octeon | |
121 | */ | |
26236fa9 RV |
122 | iq->request_list = vmalloc_node((sizeof(*iq->request_list) * num_descs), |
123 | numa_node); | |
124 | if (!iq->request_list) | |
125 | iq->request_list = vmalloc(sizeof(*iq->request_list) * | |
126 | num_descs); | |
f21fb3ed RV |
127 | if (!iq->request_list) { |
128 | lio_dma_free(oct, q_size, iq->base_addr, iq->base_addr_dma); | |
129 | dev_err(&oct->pci_dev->dev, "Alloc failed for IQ[%d] nr free list\n", | |
130 | iq_no); | |
131 | return 1; | |
132 | } | |
133 | ||
134 | memset(iq->request_list, 0, sizeof(*iq->request_list) * num_descs); | |
135 | ||
136 | dev_dbg(&oct->pci_dev->dev, "IQ[%d]: base: %p basedma: %llx count: %d\n", | |
137 | iq_no, iq->base_addr, iq->base_addr_dma, iq->max_count); | |
138 | ||
26236fa9 | 139 | iq->txpciq.u64 = txpciq.u64; |
f21fb3ed RV |
140 | iq->fill_threshold = (u32)conf->db_min; |
141 | iq->fill_cnt = 0; | |
142 | iq->host_write_index = 0; | |
143 | iq->octeon_read_index = 0; | |
144 | iq->flush_index = 0; | |
145 | iq->last_db_time = 0; | |
146 | iq->do_auto_flush = 1; | |
147 | iq->db_timeout = (u32)conf->db_timeout; | |
148 | atomic_set(&iq->instr_pending, 0); | |
149 | ||
150 | /* Initialize the spinlock for this instruction queue */ | |
151 | spin_lock_init(&iq->lock); | |
152 | ||
63da8404 | 153 | oct->io_qmask.iq |= (1ULL << iq_no); |
f21fb3ed RV |
154 | |
155 | /* Set the 32B/64B mode for each input queue */ | |
156 | oct->io_qmask.iq64B |= ((conf->instr_type == 64) << iq_no); | |
157 | iq->iqcmd_64B = (conf->instr_type == 64); | |
158 | ||
159 | oct->fn_list.setup_iq_regs(oct, iq_no); | |
160 | ||
aaa76724 BS |
161 | oct->check_db_wq[iq_no].wq = alloc_workqueue("check_iq_db", |
162 | WQ_MEM_RECLAIM, | |
163 | 0); | |
f21fb3ed RV |
164 | if (!oct->check_db_wq[iq_no].wq) { |
165 | lio_dma_free(oct, q_size, iq->base_addr, iq->base_addr_dma); | |
166 | dev_err(&oct->pci_dev->dev, "check db wq create failed for iq %d\n", | |
167 | iq_no); | |
168 | return 1; | |
169 | } | |
170 | ||
171 | db_wq = &oct->check_db_wq[iq_no]; | |
172 | ||
173 | INIT_DELAYED_WORK(&db_wq->wk.work, check_db_timeout); | |
174 | db_wq->wk.ctxptr = oct; | |
175 | db_wq->wk.ctxul = iq_no; | |
176 | queue_delayed_work(db_wq->wq, &db_wq->wk.work, msecs_to_jiffies(1)); | |
177 | ||
178 | return 0; | |
179 | } | |
180 | ||
181 | int octeon_delete_instr_queue(struct octeon_device *oct, u32 iq_no) | |
182 | { | |
183 | u64 desc_size = 0, q_size; | |
184 | struct octeon_instr_queue *iq = oct->instr_queue[iq_no]; | |
185 | ||
186 | cancel_delayed_work_sync(&oct->check_db_wq[iq_no].wk.work); | |
f21fb3ed RV |
187 | destroy_workqueue(oct->check_db_wq[iq_no].wq); |
188 | ||
189 | if (OCTEON_CN6XXX(oct)) | |
190 | desc_size = | |
191 | CFG_GET_IQ_INSTR_TYPE(CHIP_FIELD(oct, cn6xxx, conf)); | |
192 | ||
9686f310 | 193 | vfree(iq->request_list); |
f21fb3ed RV |
194 | |
195 | if (iq->base_addr) { | |
196 | q_size = iq->max_count * desc_size; | |
197 | lio_dma_free(oct, (u32)q_size, iq->base_addr, | |
198 | iq->base_addr_dma); | |
199 | return 0; | |
200 | } | |
201 | return 1; | |
202 | } | |
203 | ||
204 | /* Return 0 on success, 1 on failure */ | |
205 | int octeon_setup_iq(struct octeon_device *oct, | |
0cece6c5 RV |
206 | int ifidx, |
207 | int q_index, | |
26236fa9 | 208 | union oct_txpciq txpciq, |
f21fb3ed RV |
209 | u32 num_descs, |
210 | void *app_ctx) | |
211 | { | |
26236fa9 RV |
212 | u32 iq_no = (u32)txpciq.s.q_no; |
213 | int numa_node = cpu_to_node(iq_no % num_online_cpus()); | |
214 | ||
f21fb3ed RV |
215 | if (oct->instr_queue[iq_no]) { |
216 | dev_dbg(&oct->pci_dev->dev, "IQ is in use. Cannot create the IQ: %d again\n", | |
217 | iq_no); | |
26236fa9 | 218 | oct->instr_queue[iq_no]->txpciq.u64 = txpciq.u64; |
f21fb3ed RV |
219 | oct->instr_queue[iq_no]->app_ctx = app_ctx; |
220 | return 0; | |
221 | } | |
222 | oct->instr_queue[iq_no] = | |
26236fa9 RV |
223 | vmalloc_node(sizeof(struct octeon_instr_queue), numa_node); |
224 | if (!oct->instr_queue[iq_no]) | |
225 | oct->instr_queue[iq_no] = | |
226 | vmalloc(sizeof(struct octeon_instr_queue)); | |
f21fb3ed RV |
227 | if (!oct->instr_queue[iq_no]) |
228 | return 1; | |
229 | ||
230 | memset(oct->instr_queue[iq_no], 0, | |
231 | sizeof(struct octeon_instr_queue)); | |
232 | ||
0cece6c5 | 233 | oct->instr_queue[iq_no]->q_index = q_index; |
f21fb3ed | 234 | oct->instr_queue[iq_no]->app_ctx = app_ctx; |
0cece6c5 RV |
235 | oct->instr_queue[iq_no]->ifidx = ifidx; |
236 | ||
26236fa9 | 237 | if (octeon_init_instr_queue(oct, txpciq, num_descs)) { |
f21fb3ed RV |
238 | vfree(oct->instr_queue[iq_no]); |
239 | oct->instr_queue[iq_no] = NULL; | |
240 | return 1; | |
241 | } | |
242 | ||
243 | oct->num_iqs++; | |
244 | oct->fn_list.enable_io_queues(oct); | |
245 | return 0; | |
246 | } | |
247 | ||
248 | int lio_wait_for_instr_fetch(struct octeon_device *oct) | |
249 | { | |
250 | int i, retry = 1000, pending, instr_cnt = 0; | |
251 | ||
252 | do { | |
253 | instr_cnt = 0; | |
254 | ||
255 | /*for (i = 0; i < oct->num_iqs; i++) {*/ | |
63da8404 RV |
256 | for (i = 0; i < MAX_OCTEON_INSTR_QUEUES(oct); i++) { |
257 | if (!(oct->io_qmask.iq & (1ULL << i))) | |
f21fb3ed RV |
258 | continue; |
259 | pending = | |
260 | atomic_read(&oct-> | |
261 | instr_queue[i]->instr_pending); | |
262 | if (pending) | |
263 | __check_db_timeout(oct, i); | |
264 | instr_cnt += pending; | |
265 | } | |
266 | ||
267 | if (instr_cnt == 0) | |
268 | break; | |
269 | ||
270 | schedule_timeout_uninterruptible(1); | |
271 | ||
272 | } while (retry-- && instr_cnt); | |
273 | ||
274 | return instr_cnt; | |
275 | } | |
276 | ||
277 | static inline void | |
278 | ring_doorbell(struct octeon_device *oct, struct octeon_instr_queue *iq) | |
279 | { | |
280 | if (atomic_read(&oct->status) == OCT_DEV_RUNNING) { | |
281 | writel(iq->fill_cnt, iq->doorbell_reg); | |
282 | /* make sure doorbell write goes through */ | |
283 | mmiowb(); | |
284 | iq->fill_cnt = 0; | |
285 | iq->last_db_time = jiffies; | |
286 | return; | |
287 | } | |
288 | } | |
289 | ||
290 | static inline void __copy_cmd_into_iq(struct octeon_instr_queue *iq, | |
291 | u8 *cmd) | |
292 | { | |
293 | u8 *iqptr, cmdsize; | |
294 | ||
295 | cmdsize = ((iq->iqcmd_64B) ? 64 : 32); | |
296 | iqptr = iq->base_addr + (cmdsize * iq->host_write_index); | |
297 | ||
298 | memcpy(iqptr, cmd, cmdsize); | |
299 | } | |
300 | ||
301 | static inline int | |
302 | __post_command(struct octeon_device *octeon_dev __attribute__((unused)), | |
303 | struct octeon_instr_queue *iq, | |
304 | u32 force_db __attribute__((unused)), u8 *cmd) | |
305 | { | |
306 | u32 index = -1; | |
307 | ||
308 | /* This ensures that the read index does not wrap around to the same | |
309 | * position if queue gets full before Octeon could fetch any instr. | |
310 | */ | |
311 | if (atomic_read(&iq->instr_pending) >= (s32)(iq->max_count - 1)) | |
312 | return -1; | |
313 | ||
314 | __copy_cmd_into_iq(iq, cmd); | |
315 | ||
316 | /* "index" is returned, host_write_index is modified. */ | |
317 | index = iq->host_write_index; | |
318 | INCR_INDEX_BY1(iq->host_write_index, iq->max_count); | |
319 | iq->fill_cnt++; | |
320 | ||
321 | /* Flush the command into memory. We need to be sure the data is in | |
322 | * memory before indicating that the instruction is pending. | |
323 | */ | |
324 | wmb(); | |
325 | ||
326 | atomic_inc(&iq->instr_pending); | |
327 | ||
328 | return index; | |
329 | } | |
330 | ||
331 | static inline struct iq_post_status | |
332 | __post_command2(struct octeon_device *octeon_dev __attribute__((unused)), | |
333 | struct octeon_instr_queue *iq, | |
334 | u32 force_db __attribute__((unused)), u8 *cmd) | |
335 | { | |
336 | struct iq_post_status st; | |
337 | ||
338 | st.status = IQ_SEND_OK; | |
339 | ||
340 | /* This ensures that the read index does not wrap around to the same | |
341 | * position if queue gets full before Octeon could fetch any instr. | |
342 | */ | |
343 | if (atomic_read(&iq->instr_pending) >= (s32)(iq->max_count - 1)) { | |
344 | st.status = IQ_SEND_FAILED; | |
345 | st.index = -1; | |
346 | return st; | |
347 | } | |
348 | ||
349 | if (atomic_read(&iq->instr_pending) >= (s32)(iq->max_count - 2)) | |
350 | st.status = IQ_SEND_STOP; | |
351 | ||
352 | __copy_cmd_into_iq(iq, cmd); | |
353 | ||
354 | /* "index" is returned, host_write_index is modified. */ | |
355 | st.index = iq->host_write_index; | |
356 | INCR_INDEX_BY1(iq->host_write_index, iq->max_count); | |
357 | iq->fill_cnt++; | |
358 | ||
359 | /* Flush the command into memory. We need to be sure the data is in | |
360 | * memory before indicating that the instruction is pending. | |
361 | */ | |
362 | wmb(); | |
363 | ||
364 | atomic_inc(&iq->instr_pending); | |
365 | ||
366 | return st; | |
367 | } | |
368 | ||
369 | int | |
370 | octeon_register_reqtype_free_fn(struct octeon_device *oct, int reqtype, | |
371 | void (*fn)(void *)) | |
372 | { | |
373 | if (reqtype > REQTYPE_LAST) { | |
374 | dev_err(&oct->pci_dev->dev, "%s: Invalid reqtype: %d\n", | |
375 | __func__, reqtype); | |
376 | return -EINVAL; | |
377 | } | |
378 | ||
379 | reqtype_free_fn[oct->octeon_id][reqtype] = fn; | |
380 | ||
381 | return 0; | |
382 | } | |
383 | ||
384 | static inline void | |
385 | __add_to_request_list(struct octeon_instr_queue *iq, | |
386 | int idx, void *buf, int reqtype) | |
387 | { | |
388 | iq->request_list[idx].buf = buf; | |
389 | iq->request_list[idx].reqtype = reqtype; | |
390 | } | |
391 | ||
392 | int | |
393 | lio_process_iq_request_list(struct octeon_device *oct, | |
394 | struct octeon_instr_queue *iq) | |
395 | { | |
396 | int reqtype; | |
397 | void *buf; | |
398 | u32 old = iq->flush_index; | |
399 | u32 inst_count = 0; | |
400 | unsigned pkts_compl = 0, bytes_compl = 0; | |
401 | struct octeon_soft_command *sc; | |
402 | struct octeon_instr_irh *irh; | |
403 | ||
404 | while (old != iq->octeon_read_index) { | |
405 | reqtype = iq->request_list[old].reqtype; | |
406 | buf = iq->request_list[old].buf; | |
407 | ||
408 | if (reqtype == REQTYPE_NONE) | |
409 | goto skip_this; | |
410 | ||
411 | octeon_update_tx_completion_counters(buf, reqtype, &pkts_compl, | |
412 | &bytes_compl); | |
413 | ||
414 | switch (reqtype) { | |
415 | case REQTYPE_NORESP_NET: | |
416 | case REQTYPE_NORESP_NET_SG: | |
417 | case REQTYPE_RESP_NET_SG: | |
418 | reqtype_free_fn[oct->octeon_id][reqtype](buf); | |
419 | break; | |
420 | case REQTYPE_RESP_NET: | |
421 | case REQTYPE_SOFT_COMMAND: | |
422 | sc = buf; | |
423 | ||
6a885b60 | 424 | irh = (struct octeon_instr_irh *)&sc->cmd.cmd2.irh; |
f21fb3ed RV |
425 | if (irh->rflag) { |
426 | /* We're expecting a response from Octeon. | |
427 | * It's up to lio_process_ordered_list() to | |
428 | * process sc. Add sc to the ordered soft | |
429 | * command response list because we expect | |
430 | * a response from Octeon. | |
431 | */ | |
432 | spin_lock_bh(&oct->response_list | |
433 | [OCTEON_ORDERED_SC_LIST].lock); | |
434 | atomic_inc(&oct->response_list | |
435 | [OCTEON_ORDERED_SC_LIST]. | |
436 | pending_req_count); | |
437 | list_add_tail(&sc->node, &oct->response_list | |
438 | [OCTEON_ORDERED_SC_LIST].head); | |
439 | spin_unlock_bh(&oct->response_list | |
440 | [OCTEON_ORDERED_SC_LIST].lock); | |
441 | } else { | |
442 | if (sc->callback) { | |
443 | sc->callback(oct, OCTEON_REQUEST_DONE, | |
444 | sc->callback_arg); | |
445 | } | |
446 | } | |
447 | break; | |
448 | default: | |
449 | dev_err(&oct->pci_dev->dev, | |
450 | "%s Unknown reqtype: %d buf: %p at idx %d\n", | |
451 | __func__, reqtype, buf, old); | |
452 | } | |
453 | ||
454 | iq->request_list[old].buf = NULL; | |
455 | iq->request_list[old].reqtype = 0; | |
456 | ||
457 | skip_this: | |
458 | inst_count++; | |
459 | INCR_INDEX_BY1(old, iq->max_count); | |
460 | } | |
461 | if (bytes_compl) | |
462 | octeon_report_tx_completion_to_bql(iq->app_ctx, pkts_compl, | |
463 | bytes_compl); | |
464 | iq->flush_index = old; | |
465 | ||
466 | return inst_count; | |
467 | } | |
468 | ||
469 | static inline void | |
470 | update_iq_indices(struct octeon_device *oct, struct octeon_instr_queue *iq) | |
471 | { | |
472 | u32 inst_processed = 0; | |
473 | ||
474 | /* Calculate how many commands Octeon has read and move the read index | |
475 | * accordingly. | |
476 | */ | |
477 | iq->octeon_read_index = oct->fn_list.update_iq_read_idx(oct, iq); | |
478 | ||
479 | /* Move the NORESPONSE requests to the per-device completion list. */ | |
480 | if (iq->flush_index != iq->octeon_read_index) | |
481 | inst_processed = lio_process_iq_request_list(oct, iq); | |
482 | ||
5b173cf9 | 483 | if (inst_processed) { |
f21fb3ed RV |
484 | atomic_sub(inst_processed, &iq->instr_pending); |
485 | iq->stats.instr_processed += inst_processed; | |
5b173cf9 | 486 | } |
f21fb3ed RV |
487 | } |
488 | ||
489 | static void | |
490 | octeon_flush_iq(struct octeon_device *oct, struct octeon_instr_queue *iq, | |
491 | u32 pending_thresh) | |
492 | { | |
493 | if (atomic_read(&iq->instr_pending) >= (s32)pending_thresh) { | |
494 | spin_lock_bh(&iq->lock); | |
495 | update_iq_indices(oct, iq); | |
496 | spin_unlock_bh(&iq->lock); | |
497 | } | |
498 | } | |
499 | ||
500 | static void __check_db_timeout(struct octeon_device *oct, unsigned long iq_no) | |
501 | { | |
502 | struct octeon_instr_queue *iq; | |
503 | u64 next_time; | |
504 | ||
505 | if (!oct) | |
506 | return; | |
507 | iq = oct->instr_queue[iq_no]; | |
508 | if (!iq) | |
509 | return; | |
510 | ||
511 | /* If jiffies - last_db_time < db_timeout do nothing */ | |
512 | next_time = iq->last_db_time + iq->db_timeout; | |
513 | if (!time_after(jiffies, (unsigned long)next_time)) | |
514 | return; | |
515 | iq->last_db_time = jiffies; | |
516 | ||
517 | /* Get the lock and prevent tasklets. This routine gets called from | |
518 | * the poll thread. Instructions can now be posted in tasklet context | |
519 | */ | |
520 | spin_lock_bh(&iq->lock); | |
521 | if (iq->fill_cnt != 0) | |
522 | ring_doorbell(oct, iq); | |
523 | ||
524 | spin_unlock_bh(&iq->lock); | |
525 | ||
526 | /* Flush the instruction queue */ | |
527 | if (iq->do_auto_flush) | |
528 | octeon_flush_iq(oct, iq, 1); | |
529 | } | |
530 | ||
531 | /* Called by the Poll thread at regular intervals to check the instruction | |
532 | * queue for commands to be posted and for commands that were fetched by Octeon. | |
533 | */ | |
534 | static void check_db_timeout(struct work_struct *work) | |
535 | { | |
536 | struct cavium_wk *wk = (struct cavium_wk *)work; | |
537 | struct octeon_device *oct = (struct octeon_device *)wk->ctxptr; | |
538 | unsigned long iq_no = wk->ctxul; | |
539 | struct cavium_wq *db_wq = &oct->check_db_wq[iq_no]; | |
540 | ||
541 | __check_db_timeout(oct, iq_no); | |
542 | queue_delayed_work(db_wq->wq, &db_wq->wk.work, msecs_to_jiffies(1)); | |
543 | } | |
544 | ||
545 | int | |
546 | octeon_send_command(struct octeon_device *oct, u32 iq_no, | |
547 | u32 force_db, void *cmd, void *buf, | |
548 | u32 datasize, u32 reqtype) | |
549 | { | |
550 | struct iq_post_status st; | |
551 | struct octeon_instr_queue *iq = oct->instr_queue[iq_no]; | |
552 | ||
553 | spin_lock_bh(&iq->lock); | |
554 | ||
555 | st = __post_command2(oct, iq, force_db, cmd); | |
556 | ||
557 | if (st.status != IQ_SEND_FAILED) { | |
558 | octeon_report_sent_bytes_to_bql(buf, reqtype); | |
559 | __add_to_request_list(iq, st.index, buf, reqtype); | |
560 | INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, bytes_sent, datasize); | |
561 | INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, instr_posted, 1); | |
562 | ||
563 | if (iq->fill_cnt >= iq->fill_threshold || force_db) | |
564 | ring_doorbell(oct, iq); | |
565 | } else { | |
566 | INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, instr_dropped, 1); | |
567 | } | |
568 | ||
569 | spin_unlock_bh(&iq->lock); | |
570 | ||
571 | if (iq->do_auto_flush) | |
572 | octeon_flush_iq(oct, iq, 2); | |
573 | ||
574 | return st.status; | |
575 | } | |
576 | ||
577 | void | |
578 | octeon_prepare_soft_command(struct octeon_device *oct, | |
579 | struct octeon_soft_command *sc, | |
580 | u8 opcode, | |
581 | u8 subcode, | |
582 | u32 irh_ossp, | |
583 | u64 ossp0, | |
584 | u64 ossp1) | |
585 | { | |
586 | struct octeon_config *oct_cfg; | |
6a885b60 | 587 | struct octeon_instr_ih2 *ih2; |
f21fb3ed RV |
588 | struct octeon_instr_irh *irh; |
589 | struct octeon_instr_rdp *rdp; | |
590 | ||
591 | BUG_ON(opcode > 15); | |
592 | BUG_ON(subcode > 127); | |
593 | ||
594 | oct_cfg = octeon_get_conf(oct); | |
595 | ||
6a885b60 RV |
596 | ih2 = (struct octeon_instr_ih2 *)&sc->cmd.cmd2.ih2; |
597 | ih2->tagtype = ATOMIC_TAG; | |
598 | ih2->tag = LIO_CONTROL; | |
599 | ih2->raw = 1; | |
600 | ih2->grp = CFG_GET_CTRL_Q_GRP(oct_cfg); | |
f21fb3ed RV |
601 | |
602 | if (sc->datasize) { | |
6a885b60 RV |
603 | ih2->dlengsz = sc->datasize; |
604 | ih2->rs = 1; | |
f21fb3ed RV |
605 | } |
606 | ||
6a885b60 | 607 | irh = (struct octeon_instr_irh *)&sc->cmd.cmd2.irh; |
f21fb3ed RV |
608 | irh->opcode = opcode; |
609 | irh->subcode = subcode; | |
610 | ||
611 | /* opcode/subcode specific parameters (ossp) */ | |
612 | irh->ossp = irh_ossp; | |
6a885b60 RV |
613 | sc->cmd.cmd2.ossp[0] = ossp0; |
614 | sc->cmd.cmd2.ossp[1] = ossp1; | |
f21fb3ed RV |
615 | |
616 | if (sc->rdatasize) { | |
6a885b60 | 617 | rdp = (struct octeon_instr_rdp *)&sc->cmd.cmd2.rdp; |
f21fb3ed RV |
618 | rdp->pcie_port = oct->pcie_port; |
619 | rdp->rlen = sc->rdatasize; | |
620 | ||
621 | irh->rflag = 1; | |
6a885b60 | 622 | ih2->fsz = 40; /* irh+ossp[0]+ossp[1]+rdp+rptr = 40 bytes */ |
f21fb3ed RV |
623 | } else { |
624 | irh->rflag = 0; | |
6a885b60 | 625 | ih2->fsz = 24; /* irh + ossp[0] + ossp[1] = 24 bytes */ |
f21fb3ed | 626 | } |
f21fb3ed RV |
627 | } |
628 | ||
629 | int octeon_send_soft_command(struct octeon_device *oct, | |
630 | struct octeon_soft_command *sc) | |
631 | { | |
6a885b60 | 632 | struct octeon_instr_ih2 *ih2; |
f21fb3ed RV |
633 | struct octeon_instr_irh *irh; |
634 | struct octeon_instr_rdp *rdp; | |
6a885b60 | 635 | u32 len; |
f21fb3ed | 636 | |
6a885b60 RV |
637 | ih2 = (struct octeon_instr_ih2 *)&sc->cmd.cmd2.ih2; |
638 | if (ih2->dlengsz) { | |
639 | WARN_ON(!sc->dmadptr); | |
640 | sc->cmd.cmd2.dptr = sc->dmadptr; | |
f21fb3ed | 641 | } |
6a885b60 | 642 | irh = (struct octeon_instr_irh *)&sc->cmd.cmd2.irh; |
f21fb3ed RV |
643 | if (irh->rflag) { |
644 | BUG_ON(!sc->dmarptr); | |
645 | BUG_ON(!sc->status_word); | |
646 | *sc->status_word = COMPLETION_WORD_INIT; | |
647 | ||
6a885b60 | 648 | rdp = (struct octeon_instr_rdp *)&sc->cmd.cmd2.rdp; |
f21fb3ed | 649 | |
6a885b60 | 650 | sc->cmd.cmd2.rptr = sc->dmarptr; |
f21fb3ed | 651 | } |
6a885b60 | 652 | len = (u32)ih2->dlengsz; |
f21fb3ed RV |
653 | |
654 | if (sc->wait_time) | |
655 | sc->timeout = jiffies + sc->wait_time; | |
656 | ||
6a885b60 RV |
657 | return (octeon_send_command(oct, sc->iq_no, 1, &sc->cmd, sc, |
658 | len, REQTYPE_SOFT_COMMAND)); | |
f21fb3ed RV |
659 | } |
660 | ||
661 | int octeon_setup_sc_buffer_pool(struct octeon_device *oct) | |
662 | { | |
663 | int i; | |
664 | u64 dma_addr; | |
665 | struct octeon_soft_command *sc; | |
666 | ||
667 | INIT_LIST_HEAD(&oct->sc_buf_pool.head); | |
668 | spin_lock_init(&oct->sc_buf_pool.lock); | |
669 | atomic_set(&oct->sc_buf_pool.alloc_buf_count, 0); | |
670 | ||
671 | for (i = 0; i < MAX_SOFT_COMMAND_BUFFERS; i++) { | |
672 | sc = (struct octeon_soft_command *) | |
673 | lio_dma_alloc(oct, | |
674 | SOFT_COMMAND_BUFFER_SIZE, | |
675 | (dma_addr_t *)&dma_addr); | |
676 | if (!sc) | |
677 | return 1; | |
678 | ||
679 | sc->dma_addr = dma_addr; | |
680 | sc->size = SOFT_COMMAND_BUFFER_SIZE; | |
681 | ||
682 | list_add_tail(&sc->node, &oct->sc_buf_pool.head); | |
683 | } | |
684 | ||
685 | return 0; | |
686 | } | |
687 | ||
688 | int octeon_free_sc_buffer_pool(struct octeon_device *oct) | |
689 | { | |
690 | struct list_head *tmp, *tmp2; | |
691 | struct octeon_soft_command *sc; | |
692 | ||
693 | spin_lock(&oct->sc_buf_pool.lock); | |
694 | ||
695 | list_for_each_safe(tmp, tmp2, &oct->sc_buf_pool.head) { | |
696 | list_del(tmp); | |
697 | ||
698 | sc = (struct octeon_soft_command *)tmp; | |
699 | ||
700 | lio_dma_free(oct, sc->size, sc, sc->dma_addr); | |
701 | } | |
702 | ||
703 | INIT_LIST_HEAD(&oct->sc_buf_pool.head); | |
704 | ||
705 | spin_unlock(&oct->sc_buf_pool.lock); | |
706 | ||
707 | return 0; | |
708 | } | |
709 | ||
710 | struct octeon_soft_command *octeon_alloc_soft_command(struct octeon_device *oct, | |
711 | u32 datasize, | |
712 | u32 rdatasize, | |
713 | u32 ctxsize) | |
714 | { | |
715 | u64 dma_addr; | |
716 | u32 size; | |
717 | u32 offset = sizeof(struct octeon_soft_command); | |
718 | struct octeon_soft_command *sc = NULL; | |
719 | struct list_head *tmp; | |
720 | ||
721 | BUG_ON((offset + datasize + rdatasize + ctxsize) > | |
722 | SOFT_COMMAND_BUFFER_SIZE); | |
723 | ||
724 | spin_lock(&oct->sc_buf_pool.lock); | |
725 | ||
726 | if (list_empty(&oct->sc_buf_pool.head)) { | |
727 | spin_unlock(&oct->sc_buf_pool.lock); | |
728 | return NULL; | |
729 | } | |
730 | ||
731 | list_for_each(tmp, &oct->sc_buf_pool.head) | |
732 | break; | |
733 | ||
734 | list_del(tmp); | |
735 | ||
736 | atomic_inc(&oct->sc_buf_pool.alloc_buf_count); | |
737 | ||
738 | spin_unlock(&oct->sc_buf_pool.lock); | |
739 | ||
740 | sc = (struct octeon_soft_command *)tmp; | |
741 | ||
742 | dma_addr = sc->dma_addr; | |
743 | size = sc->size; | |
744 | ||
745 | memset(sc, 0, sc->size); | |
746 | ||
747 | sc->dma_addr = dma_addr; | |
748 | sc->size = size; | |
749 | ||
750 | if (ctxsize) { | |
751 | sc->ctxptr = (u8 *)sc + offset; | |
752 | sc->ctxsize = ctxsize; | |
753 | } | |
754 | ||
755 | /* Start data at 128 byte boundary */ | |
756 | offset = (offset + ctxsize + 127) & 0xffffff80; | |
757 | ||
758 | if (datasize) { | |
759 | sc->virtdptr = (u8 *)sc + offset; | |
760 | sc->dmadptr = dma_addr + offset; | |
761 | sc->datasize = datasize; | |
762 | } | |
763 | ||
764 | /* Start rdata at 128 byte boundary */ | |
765 | offset = (offset + datasize + 127) & 0xffffff80; | |
766 | ||
767 | if (rdatasize) { | |
768 | BUG_ON(rdatasize < 16); | |
769 | sc->virtrptr = (u8 *)sc + offset; | |
770 | sc->dmarptr = dma_addr + offset; | |
771 | sc->rdatasize = rdatasize; | |
772 | sc->status_word = (u64 *)((u8 *)(sc->virtrptr) + rdatasize - 8); | |
773 | } | |
774 | ||
775 | return sc; | |
776 | } | |
777 | ||
778 | void octeon_free_soft_command(struct octeon_device *oct, | |
779 | struct octeon_soft_command *sc) | |
780 | { | |
781 | spin_lock(&oct->sc_buf_pool.lock); | |
782 | ||
783 | list_add_tail(&sc->node, &oct->sc_buf_pool.head); | |
784 | ||
785 | atomic_dec(&oct->sc_buf_pool.alloc_buf_count); | |
786 | ||
787 | spin_unlock(&oct->sc_buf_pool.lock); | |
788 | } |