1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2016 Cavium, Inc.
7 #include "cptvf_algs.h"
8 #include "request_manager.h"
11 * get_free_pending_entry - get free entry from pending queue
12 * @param pqinfo: pending_qinfo structure
13 * @param qno: queue number
15 static struct pending_entry *get_free_pending_entry(struct pending_queue *q,
18 struct pending_entry *ent = NULL;
20 ent = &q->head[q->rear];
21 if (unlikely(ent->busy)) {
27 if (unlikely(q->rear == qlen))
34 static inline void pending_queue_inc_front(struct pending_qinfo *pqinfo,
37 struct pending_queue *queue = &pqinfo->queue[qno];
40 if (unlikely(queue->front == pqinfo->qlen))
44 static int setup_sgio_components(struct cpt_vf *cptvf, struct buf_ptr *list,
45 int buf_count, u8 *buffer)
49 struct sglist_component *sg_ptr = NULL;
50 struct pci_dev *pdev = cptvf->pdev;
52 if (unlikely(!list)) {
53 dev_err(&pdev->dev, "Input List pointer is NULL\n");
57 for (i = 0; i < buf_count; i++) {
58 if (likely(list[i].vptr)) {
59 list[i].dma_addr = dma_map_single(&pdev->dev,
63 if (unlikely(dma_mapping_error(&pdev->dev,
65 dev_err(&pdev->dev, "DMA map kernel buffer failed for component: %d\n",
73 components = buf_count / 4;
74 sg_ptr = (struct sglist_component *)buffer;
75 for (i = 0; i < components; i++) {
76 sg_ptr->u.s.len0 = cpu_to_be16(list[i * 4 + 0].size);
77 sg_ptr->u.s.len1 = cpu_to_be16(list[i * 4 + 1].size);
78 sg_ptr->u.s.len2 = cpu_to_be16(list[i * 4 + 2].size);
79 sg_ptr->u.s.len3 = cpu_to_be16(list[i * 4 + 3].size);
80 sg_ptr->ptr0 = cpu_to_be64(list[i * 4 + 0].dma_addr);
81 sg_ptr->ptr1 = cpu_to_be64(list[i * 4 + 1].dma_addr);
82 sg_ptr->ptr2 = cpu_to_be64(list[i * 4 + 2].dma_addr);
83 sg_ptr->ptr3 = cpu_to_be64(list[i * 4 + 3].dma_addr);
87 components = buf_count % 4;
91 sg_ptr->u.s.len2 = cpu_to_be16(list[i * 4 + 2].size);
92 sg_ptr->ptr2 = cpu_to_be64(list[i * 4 + 2].dma_addr);
95 sg_ptr->u.s.len1 = cpu_to_be16(list[i * 4 + 1].size);
96 sg_ptr->ptr1 = cpu_to_be64(list[i * 4 + 1].dma_addr);
99 sg_ptr->u.s.len0 = cpu_to_be16(list[i * 4 + 0].size);
100 sg_ptr->ptr0 = cpu_to_be64(list[i * 4 + 0].dma_addr);
109 for (j = 0; j < i; j++) {
110 if (list[j].dma_addr) {
111 dma_unmap_single(&pdev->dev, list[i].dma_addr,
112 list[i].size, DMA_BIDIRECTIONAL);
115 list[j].dma_addr = 0;
121 static inline int setup_sgio_list(struct cpt_vf *cptvf,
122 struct cpt_info_buffer *info,
123 struct cpt_request_info *req)
125 u16 g_sz_bytes = 0, s_sz_bytes = 0;
127 struct pci_dev *pdev = cptvf->pdev;
129 if (req->incnt > MAX_SG_IN_CNT || req->outcnt > MAX_SG_OUT_CNT) {
130 dev_err(&pdev->dev, "Request SG components are higher than supported\n");
132 goto scatter_gather_clean;
135 /* Setup gather (input) components */
136 g_sz_bytes = ((req->incnt + 3) / 4) * sizeof(struct sglist_component);
137 info->gather_components = kzalloc(g_sz_bytes, req->may_sleep ? GFP_KERNEL : GFP_ATOMIC);
138 if (!info->gather_components) {
140 goto scatter_gather_clean;
143 ret = setup_sgio_components(cptvf, req->in,
145 info->gather_components);
147 dev_err(&pdev->dev, "Failed to setup gather list\n");
149 goto scatter_gather_clean;
152 /* Setup scatter (output) components */
153 s_sz_bytes = ((req->outcnt + 3) / 4) * sizeof(struct sglist_component);
154 info->scatter_components = kzalloc(s_sz_bytes, req->may_sleep ? GFP_KERNEL : GFP_ATOMIC);
155 if (!info->scatter_components) {
157 goto scatter_gather_clean;
160 ret = setup_sgio_components(cptvf, req->out,
162 info->scatter_components);
164 dev_err(&pdev->dev, "Failed to setup gather list\n");
166 goto scatter_gather_clean;
169 /* Create and initialize DPTR */
170 info->dlen = g_sz_bytes + s_sz_bytes + SG_LIST_HDR_SIZE;
171 info->in_buffer = kzalloc(info->dlen, req->may_sleep ? GFP_KERNEL : GFP_ATOMIC);
172 if (!info->in_buffer) {
174 goto scatter_gather_clean;
177 ((__be16 *)info->in_buffer)[0] = cpu_to_be16(req->outcnt);
178 ((__be16 *)info->in_buffer)[1] = cpu_to_be16(req->incnt);
179 ((__be16 *)info->in_buffer)[2] = 0;
180 ((__be16 *)info->in_buffer)[3] = 0;
182 memcpy(&info->in_buffer[8], info->gather_components,
184 memcpy(&info->in_buffer[8 + g_sz_bytes],
185 info->scatter_components, s_sz_bytes);
187 info->dptr_baddr = dma_map_single(&pdev->dev,
188 (void *)info->in_buffer,
191 if (dma_mapping_error(&pdev->dev, info->dptr_baddr)) {
192 dev_err(&pdev->dev, "Mapping DPTR Failed %d\n", info->dlen);
194 goto scatter_gather_clean;
197 /* Create and initialize RPTR */
198 info->out_buffer = kzalloc(COMPLETION_CODE_SIZE, req->may_sleep ? GFP_KERNEL : GFP_ATOMIC);
199 if (!info->out_buffer) {
201 goto scatter_gather_clean;
204 *((u64 *)info->out_buffer) = ~((u64)COMPLETION_CODE_INIT);
205 info->alternate_caddr = (u64 *)info->out_buffer;
206 info->rptr_baddr = dma_map_single(&pdev->dev,
207 (void *)info->out_buffer,
208 COMPLETION_CODE_SIZE,
210 if (dma_mapping_error(&pdev->dev, info->rptr_baddr)) {
211 dev_err(&pdev->dev, "Mapping RPTR Failed %d\n",
212 COMPLETION_CODE_SIZE);
214 goto scatter_gather_clean;
219 scatter_gather_clean:
223 static int send_cpt_command(struct cpt_vf *cptvf, union cpt_inst_s *cmd,
226 struct pci_dev *pdev = cptvf->pdev;
227 struct command_qinfo *qinfo = NULL;
228 struct command_queue *queue;
229 struct command_chunk *chunk;
233 if (unlikely(qno >= cptvf->nr_queues)) {
234 dev_err(&pdev->dev, "Invalid queue (qno: %d, nr_queues: %d)\n",
235 qno, cptvf->nr_queues);
239 qinfo = &cptvf->cqinfo;
240 queue = &qinfo->queue[qno];
241 /* lock commad queue */
242 spin_lock(&queue->lock);
243 ent = &queue->qhead->head[queue->idx * qinfo->cmd_size];
244 memcpy(ent, (void *)cmd, qinfo->cmd_size);
246 if (++queue->idx >= queue->qhead->size / 64) {
247 struct hlist_node *node;
249 hlist_for_each(node, &queue->chead) {
250 chunk = hlist_entry(node, struct command_chunk,
252 if (chunk == queue->qhead) {
255 queue->qhead = chunk;
261 /* make sure all memory stores are done before ringing doorbell */
263 cptvf_write_vq_doorbell(cptvf, 1);
264 /* unlock command queue */
265 spin_unlock(&queue->lock);
270 static void do_request_cleanup(struct cpt_vf *cptvf,
271 struct cpt_info_buffer *info)
274 struct pci_dev *pdev = cptvf->pdev;
275 struct cpt_request_info *req;
277 if (info->dptr_baddr)
278 dma_unmap_single(&pdev->dev, info->dptr_baddr,
279 info->dlen, DMA_BIDIRECTIONAL);
281 if (info->rptr_baddr)
282 dma_unmap_single(&pdev->dev, info->rptr_baddr,
283 COMPLETION_CODE_SIZE, DMA_BIDIRECTIONAL);
285 if (info->comp_baddr)
286 dma_unmap_single(&pdev->dev, info->comp_baddr,
287 sizeof(union cpt_res_s), DMA_BIDIRECTIONAL);
291 for (i = 0; i < req->outcnt; i++) {
292 if (req->out[i].dma_addr)
293 dma_unmap_single(&pdev->dev,
294 req->out[i].dma_addr,
299 for (i = 0; i < req->incnt; i++) {
300 if (req->in[i].dma_addr)
301 dma_unmap_single(&pdev->dev,
308 kfree_sensitive(info->scatter_components);
309 kfree_sensitive(info->gather_components);
310 kfree_sensitive(info->out_buffer);
311 kfree_sensitive(info->in_buffer);
312 kfree_sensitive((void *)info->completion_addr);
313 kfree_sensitive(info);
316 static void do_post_process(struct cpt_vf *cptvf, struct cpt_info_buffer *info)
318 struct pci_dev *pdev = cptvf->pdev;
321 dev_err(&pdev->dev, "incorrect cpt_info_buffer for post processing\n");
325 do_request_cleanup(cptvf, info);
328 static inline void process_pending_queue(struct cpt_vf *cptvf,
329 struct pending_qinfo *pqinfo,
332 struct pci_dev *pdev = cptvf->pdev;
333 struct pending_queue *pqueue = &pqinfo->queue[qno];
334 struct pending_entry *pentry = NULL;
335 struct cpt_info_buffer *info = NULL;
336 union cpt_res_s *status = NULL;
340 spin_lock_bh(&pqueue->lock);
341 pentry = &pqueue->head[pqueue->front];
342 if (unlikely(!pentry->busy)) {
343 spin_unlock_bh(&pqueue->lock);
347 info = (struct cpt_info_buffer *)pentry->post_arg;
348 if (unlikely(!info)) {
349 dev_err(&pdev->dev, "Pending Entry post arg NULL\n");
350 pending_queue_inc_front(pqinfo, qno);
351 spin_unlock_bh(&pqueue->lock);
355 status = (union cpt_res_s *)pentry->completion_addr;
356 ccode = status->s.compcode;
357 if ((status->s.compcode == CPT_COMP_E_FAULT) ||
358 (status->s.compcode == CPT_COMP_E_SWERR)) {
359 dev_err(&pdev->dev, "Request failed with %s\n",
360 (status->s.compcode == CPT_COMP_E_FAULT) ?
361 "DMA Fault" : "Software error");
362 pentry->completion_addr = NULL;
363 pentry->busy = false;
364 atomic64_dec((&pqueue->pending_count));
365 pentry->post_arg = NULL;
366 pending_queue_inc_front(pqinfo, qno);
367 do_request_cleanup(cptvf, info);
368 spin_unlock_bh(&pqueue->lock);
370 } else if (status->s.compcode == COMPLETION_CODE_INIT) {
371 /* check for timeout */
372 if (time_after_eq(jiffies,
374 (CPT_COMMAND_TIMEOUT * HZ)))) {
375 dev_err(&pdev->dev, "Request timed out");
376 pentry->completion_addr = NULL;
377 pentry->busy = false;
378 atomic64_dec((&pqueue->pending_count));
379 pentry->post_arg = NULL;
380 pending_queue_inc_front(pqinfo, qno);
381 do_request_cleanup(cptvf, info);
382 spin_unlock_bh(&pqueue->lock);
384 } else if ((*info->alternate_caddr ==
385 (~COMPLETION_CODE_INIT)) &&
386 (info->extra_time < TIME_IN_RESET_COUNT)) {
387 info->time_in = jiffies;
389 spin_unlock_bh(&pqueue->lock);
394 pentry->completion_addr = NULL;
395 pentry->busy = false;
396 pentry->post_arg = NULL;
397 atomic64_dec((&pqueue->pending_count));
398 pending_queue_inc_front(pqinfo, qno);
399 spin_unlock_bh(&pqueue->lock);
401 do_post_process(info->cptvf, info);
403 * Calling callback after we find
404 * that the request has been serviced
406 pentry->callback(ccode, pentry->callback_arg);
410 int process_request(struct cpt_vf *cptvf, struct cpt_request_info *req)
412 int ret = 0, clear = 0, queue = 0;
413 struct cpt_info_buffer *info = NULL;
414 struct cptvf_request *cpt_req = NULL;
415 union ctrl_info *ctrl = NULL;
416 union cpt_res_s *result = NULL;
417 struct pending_entry *pentry = NULL;
418 struct pending_queue *pqueue = NULL;
419 struct pci_dev *pdev = cptvf->pdev;
421 struct cpt_vq_command vq_cmd;
422 union cpt_inst_s cptinst;
424 info = kzalloc(sizeof(*info), req->may_sleep ? GFP_KERNEL : GFP_ATOMIC);
425 if (unlikely(!info)) {
426 dev_err(&pdev->dev, "Unable to allocate memory for info_buffer\n");
430 cpt_req = (struct cptvf_request *)&req->req;
431 ctrl = (union ctrl_info *)&req->ctrl;
435 ret = setup_sgio_list(cptvf, info, req);
437 dev_err(&pdev->dev, "Setting up SG list failed");
438 goto request_cleanup;
441 cpt_req->dlen = info->dlen;
443 * Get buffer for union cpt_res_s response
444 * structure and its physical address
446 info->completion_addr = kzalloc(sizeof(union cpt_res_s), req->may_sleep ? GFP_KERNEL : GFP_ATOMIC);
447 if (unlikely(!info->completion_addr)) {
448 dev_err(&pdev->dev, "Unable to allocate memory for completion_addr\n");
450 goto request_cleanup;
453 result = (union cpt_res_s *)info->completion_addr;
454 result->s.compcode = COMPLETION_CODE_INIT;
455 info->comp_baddr = dma_map_single(&pdev->dev,
456 (void *)info->completion_addr,
457 sizeof(union cpt_res_s),
459 if (dma_mapping_error(&pdev->dev, info->comp_baddr)) {
460 dev_err(&pdev->dev, "mapping compptr Failed %lu\n",
461 sizeof(union cpt_res_s));
463 goto request_cleanup;
466 /* Fill the VQ command */
468 vq_cmd.cmd.s.opcode = cpu_to_be16(cpt_req->opcode.flags);
469 vq_cmd.cmd.s.param1 = cpu_to_be16(cpt_req->param1);
470 vq_cmd.cmd.s.param2 = cpu_to_be16(cpt_req->param2);
471 vq_cmd.cmd.s.dlen = cpu_to_be16(cpt_req->dlen);
473 vq_cmd.dptr = info->dptr_baddr;
474 vq_cmd.rptr = info->rptr_baddr;
476 vq_cmd.cptr.s.grp = group;
477 /* Get Pending Entry to submit command */
478 /* Always queue 0, because 1 queue per VF */
480 pqueue = &cptvf->pqinfo.queue[queue];
482 if (atomic64_read(&pqueue->pending_count) > PENDING_THOLD) {
483 dev_err(&pdev->dev, "pending threshold reached\n");
484 process_pending_queue(cptvf, &cptvf->pqinfo, queue);
488 spin_lock_bh(&pqueue->lock);
489 pentry = get_free_pending_entry(pqueue, cptvf->pqinfo.qlen);
490 if (unlikely(!pentry)) {
491 spin_unlock_bh(&pqueue->lock);
493 process_pending_queue(cptvf, &cptvf->pqinfo, queue);
495 goto get_pending_entry;
497 dev_err(&pdev->dev, "Get free entry failed\n");
498 dev_err(&pdev->dev, "queue: %d, rear: %d, front: %d\n",
499 queue, pqueue->rear, pqueue->front);
501 goto request_cleanup;
504 pentry->completion_addr = info->completion_addr;
505 pentry->post_arg = (void *)info;
506 pentry->callback = req->callback;
507 pentry->callback_arg = req->callback_arg;
508 info->pentry = pentry;
510 atomic64_inc(&pqueue->pending_count);
512 /* Send CPT command */
513 info->pentry = pentry;
514 info->time_in = jiffies;
517 /* Create the CPT_INST_S type command for HW intrepretation */
518 cptinst.s.doneint = true;
519 cptinst.s.res_addr = (u64)info->comp_baddr;
522 cptinst.s.wq_ptr = 0;
523 cptinst.s.ei0 = vq_cmd.cmd.u64;
524 cptinst.s.ei1 = vq_cmd.dptr;
525 cptinst.s.ei2 = vq_cmd.rptr;
526 cptinst.s.ei3 = vq_cmd.cptr.u64;
528 ret = send_cpt_command(cptvf, &cptinst, queue);
529 spin_unlock_bh(&pqueue->lock);
531 dev_err(&pdev->dev, "Send command failed for AE\n");
533 goto request_cleanup;
539 dev_dbg(&pdev->dev, "Failed to submit CPT command\n");
540 do_request_cleanup(cptvf, info);
545 void vq_post_process(struct cpt_vf *cptvf, u32 qno)
547 struct pci_dev *pdev = cptvf->pdev;
549 if (unlikely(qno > cptvf->nr_queues)) {
550 dev_err(&pdev->dev, "Request for post processing on invalid pending queue: %u\n",
555 process_pending_queue(cptvf, &cptvf->pqinfo, qno);
558 int cptvf_do_request(void *vfdev, struct cpt_request_info *req)
560 struct cpt_vf *cptvf = (struct cpt_vf *)vfdev;
561 struct pci_dev *pdev = cptvf->pdev;
563 if (!cpt_device_ready(cptvf)) {
564 dev_err(&pdev->dev, "CPT Device is not ready");
568 if ((cptvf->vftype == SE_TYPES) && (!req->ctrl.s.se_req)) {
569 dev_err(&pdev->dev, "CPTVF-%d of SE TYPE got AE request",
572 } else if ((cptvf->vftype == AE_TYPES) && (req->ctrl.s.se_req)) {
573 dev_err(&pdev->dev, "CPTVF-%d of AE TYPE got SE request",
578 return process_request(cptvf, req);