1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2018-2021 NXP
3 */
4
5 #include <string.h>
6
7 #include <rte_eal.h>
8 #include <rte_fslmc.h>
9 #include <rte_atomic.h>
10 #include <rte_lcore.h>
11 #include <rte_rawdev.h>
12 #include <rte_rawdev_pmd.h>
13 #include <rte_malloc.h>
14 #include <rte_ring.h>
15 #include <rte_mempool.h>
16 #include <rte_prefetch.h>
17 #include <rte_kvargs.h>
18
19 #include <mc/fsl_dpdmai.h>
20 #include <portal/dpaa2_hw_pvt.h>
21 #include <portal/dpaa2_hw_dpio.h>
22
23 #include "rte_pmd_dpaa2_qdma.h"
24 #include "dpaa2_qdma.h"
25 #include "dpaa2_qdma_logs.h"
26
27 #define DPAA2_QDMA_NO_PREFETCH "no_prefetch"
28
29 /* Dynamic log type identifier */
30 int dpaa2_qdma_logtype;
31
32 uint32_t dpaa2_coherent_no_alloc_cache;
33 uint32_t dpaa2_coherent_alloc_cache;
34
35 /* QDMA device */
36 static struct qdma_device q_dev;
37
38 /* QDMA H/W queues list */
39 TAILQ_HEAD(qdma_hw_queue_list, qdma_hw_queue);
40 static struct qdma_hw_queue_list qdma_queue_list
41 = TAILQ_HEAD_INITIALIZER(qdma_queue_list);
42
43 /* QDMA per core data */
44 static struct qdma_per_core_info qdma_core_info[RTE_MAX_LCORE];
45
46 static inline int
qdma_populate_fd_pci(phys_addr_t src,phys_addr_t dest,uint32_t len,struct qbman_fd * fd,struct rte_qdma_rbp * rbp,int ser)47 qdma_populate_fd_pci(phys_addr_t src, phys_addr_t dest,
48 uint32_t len, struct qbman_fd *fd,
49 struct rte_qdma_rbp *rbp, int ser)
50 {
51 fd->simple_pci.saddr_lo = lower_32_bits((uint64_t) (src));
52 fd->simple_pci.saddr_hi = upper_32_bits((uint64_t) (src));
53
54 fd->simple_pci.len_sl = len;
55
56 fd->simple_pci.bmt = 1;
57 fd->simple_pci.fmt = 3;
58 fd->simple_pci.sl = 1;
59 fd->simple_pci.ser = ser;
60
61 fd->simple_pci.sportid = rbp->sportid; /*pcie 3 */
62 fd->simple_pci.srbp = rbp->srbp;
63 if (rbp->srbp)
64 fd->simple_pci.rdttype = 0;
65 else
66 fd->simple_pci.rdttype = dpaa2_coherent_alloc_cache;
67
68 /*dest is pcie memory */
69 fd->simple_pci.dportid = rbp->dportid; /*pcie 3 */
70 fd->simple_pci.drbp = rbp->drbp;
71 if (rbp->drbp)
72 fd->simple_pci.wrttype = 0;
73 else
74 fd->simple_pci.wrttype = dpaa2_coherent_no_alloc_cache;
75
76 fd->simple_pci.daddr_lo = lower_32_bits((uint64_t) (dest));
77 fd->simple_pci.daddr_hi = upper_32_bits((uint64_t) (dest));
78
79 return 0;
80 }
81
82 static inline int
qdma_populate_fd_ddr(phys_addr_t src,phys_addr_t dest,uint32_t len,struct qbman_fd * fd,int ser)83 qdma_populate_fd_ddr(phys_addr_t src, phys_addr_t dest,
84 uint32_t len, struct qbman_fd *fd, int ser)
85 {
86 fd->simple_ddr.saddr_lo = lower_32_bits((uint64_t) (src));
87 fd->simple_ddr.saddr_hi = upper_32_bits((uint64_t) (src));
88
89 fd->simple_ddr.len = len;
90
91 fd->simple_ddr.bmt = 1;
92 fd->simple_ddr.fmt = 3;
93 fd->simple_ddr.sl = 1;
94 fd->simple_ddr.ser = ser;
95 /**
96 * src If RBP=0 {NS,RDTTYPE[3:0]}: 0_1011
97 * Coherent copy of cacheable memory,
98 * lookup in downstream cache, no allocate
99 * on miss
100 */
101 fd->simple_ddr.rns = 0;
102 fd->simple_ddr.rdttype = dpaa2_coherent_alloc_cache;
103 /**
104 * dest If RBP=0 {NS,WRTTYPE[3:0]}: 0_0111
105 * Coherent write of cacheable memory,
106 * lookup in downstream cache, no allocate on miss
107 */
108 fd->simple_ddr.wns = 0;
109 fd->simple_ddr.wrttype = dpaa2_coherent_no_alloc_cache;
110
111 fd->simple_ddr.daddr_lo = lower_32_bits((uint64_t) (dest));
112 fd->simple_ddr.daddr_hi = upper_32_bits((uint64_t) (dest));
113
114 return 0;
115 }
116
117 static void
dpaa2_qdma_populate_fle(struct qbman_fle * fle,uint64_t fle_iova,struct rte_qdma_rbp * rbp,uint64_t src,uint64_t dest,size_t len,uint32_t flags,uint32_t fmt)118 dpaa2_qdma_populate_fle(struct qbman_fle *fle,
119 uint64_t fle_iova,
120 struct rte_qdma_rbp *rbp,
121 uint64_t src, uint64_t dest,
122 size_t len, uint32_t flags, uint32_t fmt)
123 {
124 struct qdma_sdd *sdd;
125 uint64_t sdd_iova;
126
127 sdd = (struct qdma_sdd *)
128 ((uintptr_t)(uint64_t)fle - QDMA_FLE_FLE_OFFSET +
129 QDMA_FLE_SDD_OFFSET);
130 sdd_iova = fle_iova - QDMA_FLE_FLE_OFFSET + QDMA_FLE_SDD_OFFSET;
131
132 /* first frame list to source descriptor */
133 DPAA2_SET_FLE_ADDR(fle, sdd_iova);
134 DPAA2_SET_FLE_LEN(fle, (2 * (sizeof(struct qdma_sdd))));
135
136 /* source and destination descriptor */
137 if (rbp && rbp->enable) {
138 /* source */
139 sdd->read_cmd.portid = rbp->sportid;
140 sdd->rbpcmd_simple.pfid = rbp->spfid;
141 sdd->rbpcmd_simple.vfid = rbp->svfid;
142
143 if (rbp->srbp) {
144 sdd->read_cmd.rbp = rbp->srbp;
145 sdd->read_cmd.rdtype = DPAA2_RBP_MEM_RW;
146 } else {
147 sdd->read_cmd.rdtype = dpaa2_coherent_no_alloc_cache;
148 }
149 sdd++;
150 /* destination */
151 sdd->write_cmd.portid = rbp->dportid;
152 sdd->rbpcmd_simple.pfid = rbp->dpfid;
153 sdd->rbpcmd_simple.vfid = rbp->dvfid;
154
155 if (rbp->drbp) {
156 sdd->write_cmd.rbp = rbp->drbp;
157 sdd->write_cmd.wrttype = DPAA2_RBP_MEM_RW;
158 } else {
159 sdd->write_cmd.wrttype = dpaa2_coherent_alloc_cache;
160 }
161
162 } else {
163 sdd->read_cmd.rdtype = dpaa2_coherent_no_alloc_cache;
164 sdd++;
165 sdd->write_cmd.wrttype = dpaa2_coherent_alloc_cache;
166 }
167 fle++;
168 /* source frame list to source buffer */
169 if (flags & RTE_QDMA_JOB_SRC_PHY) {
170 DPAA2_SET_FLE_ADDR(fle, src);
171 #ifdef RTE_LIBRTE_DPAA2_USE_PHYS_IOVA
172 DPAA2_SET_FLE_BMT(fle);
173 #endif
174 } else {
175 DPAA2_SET_FLE_ADDR(fle, DPAA2_VADDR_TO_IOVA(src));
176 }
177 fle->word4.fmt = fmt;
178 DPAA2_SET_FLE_LEN(fle, len);
179
180 fle++;
181 /* destination frame list to destination buffer */
182 if (flags & RTE_QDMA_JOB_DEST_PHY) {
183 #ifdef RTE_LIBRTE_DPAA2_USE_PHYS_IOVA
184 DPAA2_SET_FLE_BMT(fle);
185 #endif
186 DPAA2_SET_FLE_ADDR(fle, dest);
187 } else {
188 DPAA2_SET_FLE_ADDR(fle, DPAA2_VADDR_TO_IOVA(dest));
189 }
190 fle->word4.fmt = fmt;
191 DPAA2_SET_FLE_LEN(fle, len);
192
193 /* Final bit: 1, for last frame list */
194 DPAA2_SET_FLE_FIN(fle);
195 }
196
dpdmai_dev_set_fd_us(struct qdma_virt_queue * qdma_vq,struct qbman_fd * fd,struct rte_qdma_job ** job,uint16_t nb_jobs)197 static inline int dpdmai_dev_set_fd_us(
198 struct qdma_virt_queue *qdma_vq,
199 struct qbman_fd *fd,
200 struct rte_qdma_job **job,
201 uint16_t nb_jobs)
202 {
203 struct rte_qdma_rbp *rbp = &qdma_vq->rbp;
204 struct rte_qdma_job **ppjob;
205 size_t iova;
206 int ret = 0, loop;
207 int ser = (qdma_vq->flags & RTE_QDMA_VQ_NO_RESPONSE) ?
208 0 : 1;
209
210 for (loop = 0; loop < nb_jobs; loop++) {
211 if (job[loop]->src & QDMA_RBP_UPPER_ADDRESS_MASK)
212 iova = (size_t)job[loop]->dest;
213 else
214 iova = (size_t)job[loop]->src;
215
216 /* Set the metadata */
217 job[loop]->vq_id = qdma_vq->vq_id;
218 ppjob = (struct rte_qdma_job **)DPAA2_IOVA_TO_VADDR(iova) - 1;
219 *ppjob = job[loop];
220
221 if ((rbp->drbp == 1) || (rbp->srbp == 1))
222 ret = qdma_populate_fd_pci((phys_addr_t)job[loop]->src,
223 (phys_addr_t)job[loop]->dest,
224 job[loop]->len, &fd[loop], rbp, ser);
225 else
226 ret = qdma_populate_fd_ddr((phys_addr_t)job[loop]->src,
227 (phys_addr_t)job[loop]->dest,
228 job[loop]->len, &fd[loop], ser);
229 }
230
231 return ret;
232 }
233
qdma_populate_sg_entry(struct rte_qdma_job ** jobs,struct qdma_sg_entry * src_sge,struct qdma_sg_entry * dst_sge,uint16_t nb_jobs)234 static uint32_t qdma_populate_sg_entry(
235 struct rte_qdma_job **jobs,
236 struct qdma_sg_entry *src_sge,
237 struct qdma_sg_entry *dst_sge,
238 uint16_t nb_jobs)
239 {
240 uint16_t i;
241 uint32_t total_len = 0;
242 uint64_t iova;
243
244 for (i = 0; i < nb_jobs; i++) {
245 /* source SG */
246 if (likely(jobs[i]->flags & RTE_QDMA_JOB_SRC_PHY)) {
247 src_sge->addr_lo = (uint32_t)jobs[i]->src;
248 src_sge->addr_hi = (jobs[i]->src >> 32);
249 } else {
250 iova = DPAA2_VADDR_TO_IOVA(jobs[i]->src);
251 src_sge->addr_lo = (uint32_t)iova;
252 src_sge->addr_hi = iova >> 32;
253 }
254 src_sge->data_len.data_len_sl0 = jobs[i]->len;
255 src_sge->ctrl.sl = QDMA_SG_SL_LONG;
256 src_sge->ctrl.fmt = QDMA_SG_FMT_SDB;
257 #ifdef RTE_LIBRTE_DPAA2_USE_PHYS_IOVA
258 src_sge->ctrl.bmt = QDMA_SG_BMT_ENABLE;
259 #else
260 src_sge->ctrl.bmt = QDMA_SG_BMT_DISABLE;
261 #endif
262 /* destination SG */
263 if (likely(jobs[i]->flags & RTE_QDMA_JOB_DEST_PHY)) {
264 dst_sge->addr_lo = (uint32_t)jobs[i]->dest;
265 dst_sge->addr_hi = (jobs[i]->dest >> 32);
266 } else {
267 iova = DPAA2_VADDR_TO_IOVA(jobs[i]->dest);
268 dst_sge->addr_lo = (uint32_t)iova;
269 dst_sge->addr_hi = iova >> 32;
270 }
271 dst_sge->data_len.data_len_sl0 = jobs[i]->len;
272 dst_sge->ctrl.sl = QDMA_SG_SL_LONG;
273 dst_sge->ctrl.fmt = QDMA_SG_FMT_SDB;
274 #ifdef RTE_LIBRTE_DPAA2_USE_PHYS_IOVA
275 dst_sge->ctrl.bmt = QDMA_SG_BMT_ENABLE;
276 #else
277 dst_sge->ctrl.bmt = QDMA_SG_BMT_DISABLE;
278 #endif
279 total_len += jobs[i]->len;
280
281 if (i == (nb_jobs - 1)) {
282 src_sge->ctrl.f = QDMA_SG_F;
283 dst_sge->ctrl.f = QDMA_SG_F;
284 } else {
285 src_sge->ctrl.f = 0;
286 dst_sge->ctrl.f = 0;
287 }
288 src_sge++;
289 dst_sge++;
290 }
291
292 return total_len;
293 }
294
dpdmai_dev_set_multi_fd_lf_no_rsp(struct qdma_virt_queue * qdma_vq,struct qbman_fd * fd,struct rte_qdma_job ** job,uint16_t nb_jobs)295 static inline int dpdmai_dev_set_multi_fd_lf_no_rsp(
296 struct qdma_virt_queue *qdma_vq,
297 struct qbman_fd *fd,
298 struct rte_qdma_job **job,
299 uint16_t nb_jobs)
300 {
301 struct rte_qdma_rbp *rbp = &qdma_vq->rbp;
302 struct rte_qdma_job **ppjob;
303 uint16_t i;
304 void *elem;
305 struct qbman_fle *fle;
306 uint64_t elem_iova, fle_iova;
307
308 for (i = 0; i < nb_jobs; i++) {
309 elem = job[i]->usr_elem;
310 #ifdef RTE_LIBRTE_DPAA2_USE_PHYS_IOVA
311 elem_iova = rte_mempool_virt2iova(elem);
312 #else
313 elem_iova = DPAA2_VADDR_TO_IOVA(elem);
314 #endif
315
316 ppjob = (struct rte_qdma_job **)
317 ((uintptr_t)(uint64_t)elem +
318 QDMA_FLE_SINGLE_JOB_OFFSET);
319 *ppjob = job[i];
320
321 job[i]->vq_id = qdma_vq->vq_id;
322
323 fle = (struct qbman_fle *)
324 ((uintptr_t)(uint64_t)elem + QDMA_FLE_FLE_OFFSET);
325 fle_iova = elem_iova + QDMA_FLE_FLE_OFFSET;
326
327 DPAA2_SET_FD_ADDR(&fd[i], fle_iova);
328 DPAA2_SET_FD_COMPOUND_FMT(&fd[i]);
329
330 memset(fle, 0, DPAA2_QDMA_MAX_FLE * sizeof(struct qbman_fle) +
331 DPAA2_QDMA_MAX_SDD * sizeof(struct qdma_sdd));
332
333 dpaa2_qdma_populate_fle(fle, fle_iova, rbp,
334 job[i]->src, job[i]->dest, job[i]->len,
335 job[i]->flags, QBMAN_FLE_WORD4_FMT_SBF);
336 }
337
338 return 0;
339 }
340
dpdmai_dev_set_multi_fd_lf(struct qdma_virt_queue * qdma_vq,struct qbman_fd * fd,struct rte_qdma_job ** job,uint16_t nb_jobs)341 static inline int dpdmai_dev_set_multi_fd_lf(
342 struct qdma_virt_queue *qdma_vq,
343 struct qbman_fd *fd,
344 struct rte_qdma_job **job,
345 uint16_t nb_jobs)
346 {
347 struct rte_qdma_rbp *rbp = &qdma_vq->rbp;
348 struct rte_qdma_job **ppjob;
349 uint16_t i;
350 int ret;
351 void *elem[RTE_QDMA_BURST_NB_MAX];
352 struct qbman_fle *fle;
353 uint64_t elem_iova, fle_iova;
354
355 ret = rte_mempool_get_bulk(qdma_vq->fle_pool, elem, nb_jobs);
356 if (ret) {
357 DPAA2_QDMA_DP_DEBUG("Memory alloc failed for FLE");
358 return ret;
359 }
360
361 for (i = 0; i < nb_jobs; i++) {
362 #ifdef RTE_LIBRTE_DPAA2_USE_PHYS_IOVA
363 elem_iova = rte_mempool_virt2iova(elem[i]);
364 #else
365 elem_iova = DPAA2_VADDR_TO_IOVA(elem[i]);
366 #endif
367
368 ppjob = (struct rte_qdma_job **)
369 ((uintptr_t)(uint64_t)elem[i] +
370 QDMA_FLE_SINGLE_JOB_OFFSET);
371 *ppjob = job[i];
372
373 job[i]->vq_id = qdma_vq->vq_id;
374
375 fle = (struct qbman_fle *)
376 ((uintptr_t)(uint64_t)elem[i] + QDMA_FLE_FLE_OFFSET);
377 fle_iova = elem_iova + QDMA_FLE_FLE_OFFSET;
378
379 DPAA2_SET_FD_ADDR(&fd[i], fle_iova);
380 DPAA2_SET_FD_COMPOUND_FMT(&fd[i]);
381 DPAA2_SET_FD_FRC(&fd[i], QDMA_SER_CTX);
382
383 memset(fle, 0, DPAA2_QDMA_MAX_FLE * sizeof(struct qbman_fle) +
384 DPAA2_QDMA_MAX_SDD * sizeof(struct qdma_sdd));
385
386 dpaa2_qdma_populate_fle(fle, fle_iova, rbp,
387 job[i]->src, job[i]->dest, job[i]->len,
388 job[i]->flags, QBMAN_FLE_WORD4_FMT_SBF);
389 }
390
391 return 0;
392 }
393
dpdmai_dev_set_sg_fd_lf(struct qdma_virt_queue * qdma_vq,struct qbman_fd * fd,struct rte_qdma_job ** job,uint16_t nb_jobs)394 static inline int dpdmai_dev_set_sg_fd_lf(
395 struct qdma_virt_queue *qdma_vq,
396 struct qbman_fd *fd,
397 struct rte_qdma_job **job,
398 uint16_t nb_jobs)
399 {
400 struct rte_qdma_rbp *rbp = &qdma_vq->rbp;
401 struct rte_qdma_job **ppjob;
402 void *elem;
403 struct qbman_fle *fle;
404 uint64_t elem_iova, fle_iova, src, dst;
405 int ret = 0, i;
406 struct qdma_sg_entry *src_sge, *dst_sge;
407 uint32_t len, fmt, flags;
408
409 /*
410 * Get an FLE/SDD from FLE pool.
411 * Note: IO metadata is before the FLE and SDD memory.
412 */
413 if (qdma_vq->flags & RTE_QDMA_VQ_NO_RESPONSE) {
414 elem = job[0]->usr_elem;
415 } else {
416 ret = rte_mempool_get(qdma_vq->fle_pool, &elem);
417 if (ret) {
418 DPAA2_QDMA_DP_DEBUG("Memory alloc failed for FLE");
419 return ret;
420 }
421 }
422
423 #ifdef RTE_LIBRTE_DPAA2_USE_PHYS_IOVA
424 elem_iova = rte_mempool_virt2iova(elem);
425 #else
426 elem_iova = DPAA2_VADDR_TO_IOVA(elem);
427 #endif
428
429 /* Set the metadata */
430 /* Save job context. */
431 *((uint16_t *)
432 ((uintptr_t)(uint64_t)elem + QDMA_FLE_JOB_NB_OFFSET)) = nb_jobs;
433 ppjob = (struct rte_qdma_job **)
434 ((uintptr_t)(uint64_t)elem + QDMA_FLE_SG_JOBS_OFFSET);
435 for (i = 0; i < nb_jobs; i++)
436 ppjob[i] = job[i];
437
438 ppjob[0]->vq_id = qdma_vq->vq_id;
439
440 fle = (struct qbman_fle *)
441 ((uintptr_t)(uint64_t)elem + QDMA_FLE_FLE_OFFSET);
442 fle_iova = elem_iova + QDMA_FLE_FLE_OFFSET;
443
444 DPAA2_SET_FD_ADDR(fd, fle_iova);
445 DPAA2_SET_FD_COMPOUND_FMT(fd);
446 if (!(qdma_vq->flags & RTE_QDMA_VQ_NO_RESPONSE))
447 DPAA2_SET_FD_FRC(fd, QDMA_SER_CTX);
448
449 /* Populate FLE */
450 if (likely(nb_jobs > 1)) {
451 src_sge = (struct qdma_sg_entry *)
452 ((uintptr_t)(uint64_t)elem + QDMA_FLE_SG_ENTRY_OFFSET);
453 dst_sge = src_sge + DPAA2_QDMA_MAX_SG_NB;
454 src = elem_iova + QDMA_FLE_SG_ENTRY_OFFSET;
455 dst = src +
456 DPAA2_QDMA_MAX_SG_NB * sizeof(struct qdma_sg_entry);
457 len = qdma_populate_sg_entry(job, src_sge, dst_sge, nb_jobs);
458 fmt = QBMAN_FLE_WORD4_FMT_SGE;
459 flags = RTE_QDMA_JOB_SRC_PHY | RTE_QDMA_JOB_DEST_PHY;
460 } else {
461 src = job[0]->src;
462 dst = job[0]->dest;
463 len = job[0]->len;
464 fmt = QBMAN_FLE_WORD4_FMT_SBF;
465 flags = job[0]->flags;
466 }
467
468 memset(fle, 0, DPAA2_QDMA_MAX_FLE * sizeof(struct qbman_fle) +
469 DPAA2_QDMA_MAX_SDD * sizeof(struct qdma_sdd));
470
471 dpaa2_qdma_populate_fle(fle, fle_iova, rbp,
472 src, dst, len, flags, fmt);
473
474 return 0;
475 }
476
dpdmai_dev_get_job_us(struct qdma_virt_queue * qdma_vq __rte_unused,const struct qbman_fd * fd,struct rte_qdma_job ** job,uint16_t * nb_jobs)477 static inline uint16_t dpdmai_dev_get_job_us(
478 struct qdma_virt_queue *qdma_vq __rte_unused,
479 const struct qbman_fd *fd,
480 struct rte_qdma_job **job, uint16_t *nb_jobs)
481 {
482 uint16_t vqid;
483 size_t iova;
484 struct rte_qdma_job **ppjob;
485
486 if (fd->simple_pci.saddr_hi & (QDMA_RBP_UPPER_ADDRESS_MASK >> 32))
487 iova = (size_t)(((uint64_t)fd->simple_pci.daddr_hi) << 32
488 | (uint64_t)fd->simple_pci.daddr_lo);
489 else
490 iova = (size_t)(((uint64_t)fd->simple_pci.saddr_hi) << 32
491 | (uint64_t)fd->simple_pci.saddr_lo);
492
493 ppjob = (struct rte_qdma_job **)DPAA2_IOVA_TO_VADDR(iova) - 1;
494 *job = (struct rte_qdma_job *)*ppjob;
495 (*job)->status = (fd->simple_pci.acc_err << 8) |
496 (fd->simple_pci.error);
497 vqid = (*job)->vq_id;
498 *nb_jobs = 1;
499
500 return vqid;
501 }
502
dpdmai_dev_get_single_job_lf(struct qdma_virt_queue * qdma_vq,const struct qbman_fd * fd,struct rte_qdma_job ** job,uint16_t * nb_jobs)503 static inline uint16_t dpdmai_dev_get_single_job_lf(
504 struct qdma_virt_queue *qdma_vq,
505 const struct qbman_fd *fd,
506 struct rte_qdma_job **job,
507 uint16_t *nb_jobs)
508 {
509 struct qbman_fle *fle;
510 struct rte_qdma_job **ppjob = NULL;
511 uint16_t status;
512
513 /*
514 * Fetch metadata from FLE. job and vq_id were set
515 * in metadata in the enqueue operation.
516 */
517 fle = (struct qbman_fle *)
518 DPAA2_IOVA_TO_VADDR(DPAA2_GET_FD_ADDR(fd));
519
520 *nb_jobs = 1;
521 ppjob = (struct rte_qdma_job **)((uintptr_t)(uint64_t)fle -
522 QDMA_FLE_FLE_OFFSET + QDMA_FLE_SINGLE_JOB_OFFSET);
523
524 status = (DPAA2_GET_FD_ERR(fd) << 8) | (DPAA2_GET_FD_FRC(fd) & 0xFF);
525
526 *job = *ppjob;
527 (*job)->status = status;
528
529 /* Free FLE to the pool */
530 rte_mempool_put(qdma_vq->fle_pool,
531 (void *)
532 ((uintptr_t)(uint64_t)fle - QDMA_FLE_FLE_OFFSET));
533
534 return (*job)->vq_id;
535 }
536
dpdmai_dev_get_sg_job_lf(struct qdma_virt_queue * qdma_vq,const struct qbman_fd * fd,struct rte_qdma_job ** job,uint16_t * nb_jobs)537 static inline uint16_t dpdmai_dev_get_sg_job_lf(
538 struct qdma_virt_queue *qdma_vq,
539 const struct qbman_fd *fd,
540 struct rte_qdma_job **job,
541 uint16_t *nb_jobs)
542 {
543 struct qbman_fle *fle;
544 struct rte_qdma_job **ppjob = NULL;
545 uint16_t i, status;
546
547 /*
548 * Fetch metadata from FLE. job and vq_id were set
549 * in metadata in the enqueue operation.
550 */
551 fle = (struct qbman_fle *)
552 DPAA2_IOVA_TO_VADDR(DPAA2_GET_FD_ADDR(fd));
553 *nb_jobs = *((uint16_t *)((uintptr_t)(uint64_t)fle -
554 QDMA_FLE_FLE_OFFSET + QDMA_FLE_JOB_NB_OFFSET));
555 ppjob = (struct rte_qdma_job **)((uintptr_t)(uint64_t)fle -
556 QDMA_FLE_FLE_OFFSET + QDMA_FLE_SG_JOBS_OFFSET);
557 status = (DPAA2_GET_FD_ERR(fd) << 8) | (DPAA2_GET_FD_FRC(fd) & 0xFF);
558
559 for (i = 0; i < (*nb_jobs); i++) {
560 job[i] = ppjob[i];
561 job[i]->status = status;
562 }
563
564 /* Free FLE to the pool */
565 rte_mempool_put(qdma_vq->fle_pool,
566 (void *)
567 ((uintptr_t)(uint64_t)fle - QDMA_FLE_FLE_OFFSET));
568
569 return job[0]->vq_id;
570 }
571
572 /* Function to receive a QDMA job for a given device and queue*/
573 static int
dpdmai_dev_dequeue_multijob_prefetch(struct qdma_virt_queue * qdma_vq,uint16_t * vq_id,struct rte_qdma_job ** job,uint16_t nb_jobs)574 dpdmai_dev_dequeue_multijob_prefetch(
575 struct qdma_virt_queue *qdma_vq,
576 uint16_t *vq_id,
577 struct rte_qdma_job **job,
578 uint16_t nb_jobs)
579 {
580 struct qdma_hw_queue *qdma_pq = qdma_vq->hw_queue;
581 struct dpaa2_dpdmai_dev *dpdmai_dev = qdma_pq->dpdmai_dev;
582 uint16_t rxq_id = qdma_pq->queue_id;
583
584 struct dpaa2_queue *rxq;
585 struct qbman_result *dq_storage, *dq_storage1 = NULL;
586 struct qbman_pull_desc pulldesc;
587 struct qbman_swp *swp;
588 struct queue_storage_info_t *q_storage;
589 uint32_t fqid;
590 uint8_t status, pending;
591 uint8_t num_rx = 0;
592 const struct qbman_fd *fd;
593 uint16_t vqid, num_rx_ret;
594 int ret, pull_size;
595
596 if (qdma_vq->flags & RTE_QDMA_VQ_FD_SG_FORMAT) {
597 /** Make sure there are enough space to get jobs.*/
598 if (unlikely(nb_jobs < DPAA2_QDMA_MAX_SG_NB))
599 return -EINVAL;
600 nb_jobs = 1;
601 }
602
603 if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
604 ret = dpaa2_affine_qbman_swp();
605 if (ret) {
606 DPAA2_QDMA_ERR(
607 "Failed to allocate IO portal, tid: %d\n",
608 rte_gettid());
609 return 0;
610 }
611 }
612 swp = DPAA2_PER_LCORE_PORTAL;
613
614 pull_size = (nb_jobs > dpaa2_dqrr_size) ? dpaa2_dqrr_size : nb_jobs;
615 rxq = &(dpdmai_dev->rx_queue[rxq_id]);
616 fqid = rxq->fqid;
617 q_storage = rxq->q_storage;
618
619 if (unlikely(!q_storage->active_dqs)) {
620 q_storage->toggle = 0;
621 dq_storage = q_storage->dq_storage[q_storage->toggle];
622 q_storage->last_num_pkts = pull_size;
623 qbman_pull_desc_clear(&pulldesc);
624 qbman_pull_desc_set_numframes(&pulldesc,
625 q_storage->last_num_pkts);
626 qbman_pull_desc_set_fq(&pulldesc, fqid);
627 qbman_pull_desc_set_storage(&pulldesc, dq_storage,
628 (size_t)(DPAA2_VADDR_TO_IOVA(dq_storage)), 1);
629 if (check_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index)) {
630 while (!qbman_check_command_complete(
631 get_swp_active_dqs(
632 DPAA2_PER_LCORE_DPIO->index)))
633 ;
634 clear_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index);
635 }
636 while (1) {
637 if (qbman_swp_pull(swp, &pulldesc)) {
638 DPAA2_QDMA_DP_WARN(
639 "VDQ command not issued.QBMAN busy\n");
640 /* Portal was busy, try again */
641 continue;
642 }
643 break;
644 }
645 q_storage->active_dqs = dq_storage;
646 q_storage->active_dpio_id = DPAA2_PER_LCORE_DPIO->index;
647 set_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index,
648 dq_storage);
649 }
650
651 dq_storage = q_storage->active_dqs;
652 rte_prefetch0((void *)(size_t)(dq_storage));
653 rte_prefetch0((void *)(size_t)(dq_storage + 1));
654
655 /* Prepare next pull descriptor. This will give space for the
656 * prefetching done on DQRR entries
657 */
658 q_storage->toggle ^= 1;
659 dq_storage1 = q_storage->dq_storage[q_storage->toggle];
660 qbman_pull_desc_clear(&pulldesc);
661 qbman_pull_desc_set_numframes(&pulldesc, pull_size);
662 qbman_pull_desc_set_fq(&pulldesc, fqid);
663 qbman_pull_desc_set_storage(&pulldesc, dq_storage1,
664 (size_t)(DPAA2_VADDR_TO_IOVA(dq_storage1)), 1);
665
666 /* Check if the previous issued command is completed.
667 * Also seems like the SWP is shared between the Ethernet Driver
668 * and the SEC driver.
669 */
670 while (!qbman_check_command_complete(dq_storage))
671 ;
672 if (dq_storage == get_swp_active_dqs(q_storage->active_dpio_id))
673 clear_swp_active_dqs(q_storage->active_dpio_id);
674
675 pending = 1;
676
677 do {
678 /* Loop until the dq_storage is updated with
679 * new token by QBMAN
680 */
681 while (!qbman_check_new_result(dq_storage))
682 ;
683 rte_prefetch0((void *)((size_t)(dq_storage + 2)));
684 /* Check whether Last Pull command is Expired and
685 * setting Condition for Loop termination
686 */
687 if (qbman_result_DQ_is_pull_complete(dq_storage)) {
688 pending = 0;
689 /* Check for valid frame. */
690 status = qbman_result_DQ_flags(dq_storage);
691 if (unlikely((status & QBMAN_DQ_STAT_VALIDFRAME) == 0))
692 continue;
693 }
694 fd = qbman_result_DQ_fd(dq_storage);
695
696 vqid = qdma_vq->get_job(qdma_vq, fd, &job[num_rx],
697 &num_rx_ret);
698 if (vq_id)
699 vq_id[num_rx] = vqid;
700
701 dq_storage++;
702 num_rx += num_rx_ret;
703 } while (pending);
704
705 if (check_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index)) {
706 while (!qbman_check_command_complete(
707 get_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index)))
708 ;
709 clear_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index);
710 }
711 /* issue a volatile dequeue command for next pull */
712 while (1) {
713 if (qbman_swp_pull(swp, &pulldesc)) {
714 DPAA2_QDMA_DP_WARN(
715 "VDQ command is not issued. QBMAN is busy (2)\n");
716 continue;
717 }
718 break;
719 }
720
721 q_storage->active_dqs = dq_storage1;
722 q_storage->active_dpio_id = DPAA2_PER_LCORE_DPIO->index;
723 set_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index, dq_storage1);
724
725 return num_rx;
726 }
727
728 static int
dpdmai_dev_dequeue_multijob_no_prefetch(struct qdma_virt_queue * qdma_vq,uint16_t * vq_id,struct rte_qdma_job ** job,uint16_t nb_jobs)729 dpdmai_dev_dequeue_multijob_no_prefetch(
730 struct qdma_virt_queue *qdma_vq,
731 uint16_t *vq_id,
732 struct rte_qdma_job **job,
733 uint16_t nb_jobs)
734 {
735 struct qdma_hw_queue *qdma_pq = qdma_vq->hw_queue;
736 struct dpaa2_dpdmai_dev *dpdmai_dev = qdma_pq->dpdmai_dev;
737 uint16_t rxq_id = qdma_pq->queue_id;
738
739 struct dpaa2_queue *rxq;
740 struct qbman_result *dq_storage;
741 struct qbman_pull_desc pulldesc;
742 struct qbman_swp *swp;
743 uint32_t fqid;
744 uint8_t status, pending;
745 uint8_t num_rx = 0;
746 const struct qbman_fd *fd;
747 uint16_t vqid, num_rx_ret;
748 int ret, next_pull, num_pulled = 0;
749
750 if (qdma_vq->flags & RTE_QDMA_VQ_FD_SG_FORMAT) {
751 /** Make sure there are enough space to get jobs.*/
752 if (unlikely(nb_jobs < DPAA2_QDMA_MAX_SG_NB))
753 return -EINVAL;
754 nb_jobs = 1;
755 }
756
757 next_pull = nb_jobs;
758
759 if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
760 ret = dpaa2_affine_qbman_swp();
761 if (ret) {
762 DPAA2_QDMA_ERR(
763 "Failed to allocate IO portal, tid: %d\n",
764 rte_gettid());
765 return 0;
766 }
767 }
768 swp = DPAA2_PER_LCORE_PORTAL;
769
770 rxq = &(dpdmai_dev->rx_queue[rxq_id]);
771 fqid = rxq->fqid;
772
773 do {
774 dq_storage = rxq->q_storage->dq_storage[0];
775 /* Prepare dequeue descriptor */
776 qbman_pull_desc_clear(&pulldesc);
777 qbman_pull_desc_set_fq(&pulldesc, fqid);
778 qbman_pull_desc_set_storage(&pulldesc, dq_storage,
779 (uint64_t)(DPAA2_VADDR_TO_IOVA(dq_storage)), 1);
780
781 if (next_pull > dpaa2_dqrr_size) {
782 qbman_pull_desc_set_numframes(&pulldesc,
783 dpaa2_dqrr_size);
784 next_pull -= dpaa2_dqrr_size;
785 } else {
786 qbman_pull_desc_set_numframes(&pulldesc, next_pull);
787 next_pull = 0;
788 }
789
790 while (1) {
791 if (qbman_swp_pull(swp, &pulldesc)) {
792 DPAA2_QDMA_DP_WARN(
793 "VDQ command not issued. QBMAN busy");
794 /* Portal was busy, try again */
795 continue;
796 }
797 break;
798 }
799
800 rte_prefetch0((void *)((size_t)(dq_storage + 1)));
801 /* Check if the previous issued command is completed. */
802 while (!qbman_check_command_complete(dq_storage))
803 ;
804
805 num_pulled = 0;
806 pending = 1;
807
808 do {
809 /* Loop until dq_storage is updated
810 * with new token by QBMAN
811 */
812 while (!qbman_check_new_result(dq_storage))
813 ;
814 rte_prefetch0((void *)((size_t)(dq_storage + 2)));
815
816 if (qbman_result_DQ_is_pull_complete(dq_storage)) {
817 pending = 0;
818 /* Check for valid frame. */
819 status = qbman_result_DQ_flags(dq_storage);
820 if (unlikely((status &
821 QBMAN_DQ_STAT_VALIDFRAME) == 0))
822 continue;
823 }
824 fd = qbman_result_DQ_fd(dq_storage);
825
826 vqid = qdma_vq->get_job(qdma_vq, fd,
827 &job[num_rx], &num_rx_ret);
828 if (vq_id)
829 vq_id[num_rx] = vqid;
830
831 dq_storage++;
832 num_rx += num_rx_ret;
833 num_pulled++;
834
835 } while (pending);
836 /* Last VDQ provided all packets and more packets are requested */
837 } while (next_pull && num_pulled == dpaa2_dqrr_size);
838
839 return num_rx;
840 }
841
842 static int
dpdmai_dev_enqueue_multi(struct qdma_virt_queue * qdma_vq,struct rte_qdma_job ** job,uint16_t nb_jobs)843 dpdmai_dev_enqueue_multi(
844 struct qdma_virt_queue *qdma_vq,
845 struct rte_qdma_job **job,
846 uint16_t nb_jobs)
847 {
848 struct qdma_hw_queue *qdma_pq = qdma_vq->hw_queue;
849 struct dpaa2_dpdmai_dev *dpdmai_dev = qdma_pq->dpdmai_dev;
850 uint16_t txq_id = qdma_pq->queue_id;
851
852 struct qbman_fd fd[RTE_QDMA_BURST_NB_MAX];
853 struct dpaa2_queue *txq;
854 struct qbman_eq_desc eqdesc;
855 struct qbman_swp *swp;
856 int ret;
857 uint32_t num_to_send = 0;
858 uint16_t num_tx = 0;
859 uint32_t enqueue_loop, retry_count, loop;
860
861 if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
862 ret = dpaa2_affine_qbman_swp();
863 if (ret) {
864 DPAA2_QDMA_ERR(
865 "Failed to allocate IO portal, tid: %d\n",
866 rte_gettid());
867 return 0;
868 }
869 }
870 swp = DPAA2_PER_LCORE_PORTAL;
871
872 txq = &(dpdmai_dev->tx_queue[txq_id]);
873
874 /* Prepare enqueue descriptor */
875 qbman_eq_desc_clear(&eqdesc);
876 qbman_eq_desc_set_fq(&eqdesc, txq->fqid);
877 qbman_eq_desc_set_no_orp(&eqdesc, 0);
878 qbman_eq_desc_set_response(&eqdesc, 0, 0);
879
880 if (qdma_vq->flags & RTE_QDMA_VQ_FD_SG_FORMAT) {
881 uint16_t fd_nb;
882 uint16_t sg_entry_nb = nb_jobs > DPAA2_QDMA_MAX_SG_NB ?
883 DPAA2_QDMA_MAX_SG_NB : nb_jobs;
884 uint16_t job_idx = 0;
885 uint16_t fd_sg_nb[8];
886 uint16_t nb_jobs_ret = 0;
887
888 if (nb_jobs % DPAA2_QDMA_MAX_SG_NB)
889 fd_nb = nb_jobs / DPAA2_QDMA_MAX_SG_NB + 1;
890 else
891 fd_nb = nb_jobs / DPAA2_QDMA_MAX_SG_NB;
892
893 memset(&fd[0], 0, sizeof(struct qbman_fd) * fd_nb);
894
895 for (loop = 0; loop < fd_nb; loop++) {
896 ret = qdma_vq->set_fd(qdma_vq, &fd[loop], &job[job_idx],
897 sg_entry_nb);
898 if (unlikely(ret < 0))
899 return 0;
900 fd_sg_nb[loop] = sg_entry_nb;
901 nb_jobs -= sg_entry_nb;
902 job_idx += sg_entry_nb;
903 sg_entry_nb = nb_jobs > DPAA2_QDMA_MAX_SG_NB ?
904 DPAA2_QDMA_MAX_SG_NB : nb_jobs;
905 }
906
907 /* Enqueue the packet to the QBMAN */
908 enqueue_loop = 0; retry_count = 0;
909
910 while (enqueue_loop < fd_nb) {
911 ret = qbman_swp_enqueue_multiple(swp,
912 &eqdesc, &fd[enqueue_loop],
913 NULL, fd_nb - enqueue_loop);
914 if (unlikely(ret < 0)) {
915 retry_count++;
916 if (retry_count > DPAA2_MAX_TX_RETRY_COUNT)
917 return nb_jobs_ret;
918 } else {
919 for (loop = 0; loop < (uint32_t)ret; loop++)
920 nb_jobs_ret +=
921 fd_sg_nb[enqueue_loop + loop];
922 enqueue_loop += ret;
923 retry_count = 0;
924 }
925 }
926
927 return nb_jobs_ret;
928 }
929
930 memset(fd, 0, nb_jobs * sizeof(struct qbman_fd));
931
932 while (nb_jobs > 0) {
933 num_to_send = (nb_jobs > dpaa2_eqcr_size) ?
934 dpaa2_eqcr_size : nb_jobs;
935
936 ret = qdma_vq->set_fd(qdma_vq, &fd[num_tx],
937 &job[num_tx], num_to_send);
938 if (unlikely(ret < 0))
939 break;
940
941 /* Enqueue the packet to the QBMAN */
942 enqueue_loop = 0; retry_count = 0;
943 loop = num_to_send;
944
945 while (enqueue_loop < loop) {
946 ret = qbman_swp_enqueue_multiple(swp,
947 &eqdesc,
948 &fd[num_tx + enqueue_loop],
949 NULL,
950 loop - enqueue_loop);
951 if (unlikely(ret < 0)) {
952 retry_count++;
953 if (retry_count > DPAA2_MAX_TX_RETRY_COUNT)
954 return num_tx;
955 } else {
956 enqueue_loop += ret;
957 retry_count = 0;
958 }
959 }
960 num_tx += num_to_send;
961 nb_jobs -= loop;
962 }
963 return num_tx;
964 }
965
966 static struct qdma_hw_queue *
alloc_hw_queue(uint32_t lcore_id)967 alloc_hw_queue(uint32_t lcore_id)
968 {
969 struct qdma_hw_queue *queue = NULL;
970
971 DPAA2_QDMA_FUNC_TRACE();
972
973 /* Get a free queue from the list */
974 TAILQ_FOREACH(queue, &qdma_queue_list, next) {
975 if (queue->num_users == 0) {
976 queue->lcore_id = lcore_id;
977 queue->num_users++;
978 break;
979 }
980 }
981
982 return queue;
983 }
984
985 static void
free_hw_queue(struct qdma_hw_queue * queue)986 free_hw_queue(struct qdma_hw_queue *queue)
987 {
988 DPAA2_QDMA_FUNC_TRACE();
989
990 queue->num_users--;
991 }
992
993
994 static struct qdma_hw_queue *
get_hw_queue(struct qdma_device * qdma_dev,uint32_t lcore_id)995 get_hw_queue(struct qdma_device *qdma_dev, uint32_t lcore_id)
996 {
997 struct qdma_per_core_info *core_info;
998 struct qdma_hw_queue *queue, *temp;
999 uint32_t least_num_users;
1000 int num_hw_queues, i;
1001
1002 DPAA2_QDMA_FUNC_TRACE();
1003
1004 core_info = &qdma_core_info[lcore_id];
1005 num_hw_queues = core_info->num_hw_queues;
1006
1007 /*
1008 * Allocate a HW queue if there are less queues
1009 * than maximum per core queues configured
1010 */
1011 if (num_hw_queues < qdma_dev->max_hw_queues_per_core) {
1012 queue = alloc_hw_queue(lcore_id);
1013 if (queue) {
1014 core_info->hw_queues[num_hw_queues] = queue;
1015 core_info->num_hw_queues++;
1016 return queue;
1017 }
1018 }
1019
1020 queue = core_info->hw_queues[0];
1021 /* In case there is no queue associated with the core return NULL */
1022 if (!queue)
1023 return NULL;
1024
1025 /* Fetch the least loaded H/W queue */
1026 least_num_users = core_info->hw_queues[0]->num_users;
1027 for (i = 0; i < num_hw_queues; i++) {
1028 temp = core_info->hw_queues[i];
1029 if (temp->num_users < least_num_users)
1030 queue = temp;
1031 }
1032
1033 if (queue)
1034 queue->num_users++;
1035
1036 return queue;
1037 }
1038
1039 static void
put_hw_queue(struct qdma_hw_queue * queue)1040 put_hw_queue(struct qdma_hw_queue *queue)
1041 {
1042 struct qdma_per_core_info *core_info;
1043 int lcore_id, num_hw_queues, i;
1044
1045 DPAA2_QDMA_FUNC_TRACE();
1046
1047 /*
1048 * If this is the last user of the queue free it.
1049 * Also remove it from QDMA core info.
1050 */
1051 if (queue->num_users == 1) {
1052 free_hw_queue(queue);
1053
1054 /* Remove the physical queue from core info */
1055 lcore_id = queue->lcore_id;
1056 core_info = &qdma_core_info[lcore_id];
1057 num_hw_queues = core_info->num_hw_queues;
1058 for (i = 0; i < num_hw_queues; i++) {
1059 if (queue == core_info->hw_queues[i])
1060 break;
1061 }
1062 for (; i < num_hw_queues - 1; i++)
1063 core_info->hw_queues[i] = core_info->hw_queues[i + 1];
1064 core_info->hw_queues[i] = NULL;
1065 } else {
1066 queue->num_users--;
1067 }
1068 }
1069
1070 static int
dpaa2_qdma_attr_get(struct rte_rawdev * rawdev,__rte_unused const char * attr_name,uint64_t * attr_value)1071 dpaa2_qdma_attr_get(struct rte_rawdev *rawdev,
1072 __rte_unused const char *attr_name,
1073 uint64_t *attr_value)
1074 {
1075 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1076 struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
1077 struct rte_qdma_attr *qdma_attr = (struct rte_qdma_attr *)attr_value;
1078
1079 DPAA2_QDMA_FUNC_TRACE();
1080
1081 qdma_attr->num_hw_queues = qdma_dev->num_hw_queues;
1082
1083 return 0;
1084 }
1085
1086 static int
dpaa2_qdma_reset(struct rte_rawdev * rawdev)1087 dpaa2_qdma_reset(struct rte_rawdev *rawdev)
1088 {
1089 struct qdma_hw_queue *queue;
1090 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1091 struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
1092 int i;
1093
1094 DPAA2_QDMA_FUNC_TRACE();
1095
1096 /* In case QDMA device is not in stopped state, return -EBUSY */
1097 if (qdma_dev->state == 1) {
1098 DPAA2_QDMA_ERR(
1099 "Device is in running state. Stop before reset.");
1100 return -EBUSY;
1101 }
1102
1103 /* In case there are pending jobs on any VQ, return -EBUSY */
1104 for (i = 0; i < qdma_dev->max_vqs; i++) {
1105 if (qdma_dev->vqs[i].in_use && (qdma_dev->vqs[i].num_enqueues !=
1106 qdma_dev->vqs[i].num_dequeues)) {
1107 DPAA2_QDMA_ERR("Jobs are still pending on VQ: %d", i);
1108 return -EBUSY;
1109 }
1110 }
1111
1112 /* Reset HW queues */
1113 TAILQ_FOREACH(queue, &qdma_queue_list, next)
1114 queue->num_users = 0;
1115
1116 /* Reset and free virtual queues */
1117 for (i = 0; i < qdma_dev->max_vqs; i++) {
1118 rte_ring_free(qdma_dev->vqs[i].status_ring);
1119 }
1120 rte_free(qdma_dev->vqs);
1121 qdma_dev->vqs = NULL;
1122
1123 /* Reset per core info */
1124 memset(&qdma_core_info, 0,
1125 sizeof(struct qdma_per_core_info) * RTE_MAX_LCORE);
1126
1127 /* Reset QDMA device structure */
1128 qdma_dev->max_hw_queues_per_core = 0;
1129 qdma_dev->fle_queue_pool_cnt = 0;
1130 qdma_dev->max_vqs = 0;
1131
1132 return 0;
1133 }
1134
1135 static int
dpaa2_qdma_configure(const struct rte_rawdev * rawdev,rte_rawdev_obj_t config,size_t config_size)1136 dpaa2_qdma_configure(const struct rte_rawdev *rawdev,
1137 rte_rawdev_obj_t config,
1138 size_t config_size)
1139 {
1140 char name[32]; /* RTE_MEMZONE_NAMESIZE = 32 */
1141 struct rte_qdma_config *qdma_config = (struct rte_qdma_config *)config;
1142 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1143 struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
1144
1145 DPAA2_QDMA_FUNC_TRACE();
1146
1147 if (config_size != sizeof(*qdma_config)) {
1148 DPAA2_QDMA_ERR("Config size mismatch. Expected %" PRIu64
1149 ", Got: %" PRIu64, (uint64_t)(sizeof(*qdma_config)),
1150 (uint64_t)config_size);
1151 return -EINVAL;
1152 }
1153
1154 /* In case QDMA device is not in stopped state, return -EBUSY */
1155 if (qdma_dev->state == 1) {
1156 DPAA2_QDMA_ERR(
1157 "Device is in running state. Stop before config.");
1158 return -1;
1159 }
1160
1161 /* Set max HW queue per core */
1162 if (qdma_config->max_hw_queues_per_core > MAX_HW_QUEUE_PER_CORE) {
1163 DPAA2_QDMA_ERR("H/W queues per core is more than: %d",
1164 MAX_HW_QUEUE_PER_CORE);
1165 return -EINVAL;
1166 }
1167 qdma_dev->max_hw_queues_per_core =
1168 qdma_config->max_hw_queues_per_core;
1169
1170 /* Allocate Virtual Queues */
1171 sprintf(name, "qdma_%d_vq", rawdev->dev_id);
1172 qdma_dev->vqs = rte_malloc(name,
1173 (sizeof(struct qdma_virt_queue) * qdma_config->max_vqs),
1174 RTE_CACHE_LINE_SIZE);
1175 if (!qdma_dev->vqs) {
1176 DPAA2_QDMA_ERR("qdma_virtual_queues allocation failed");
1177 return -ENOMEM;
1178 }
1179 qdma_dev->max_vqs = qdma_config->max_vqs;
1180 qdma_dev->fle_queue_pool_cnt = qdma_config->fle_queue_pool_cnt;
1181
1182 return 0;
1183 }
1184
1185 static int
dpaa2_qdma_start(struct rte_rawdev * rawdev)1186 dpaa2_qdma_start(struct rte_rawdev *rawdev)
1187 {
1188 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1189 struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
1190
1191 DPAA2_QDMA_FUNC_TRACE();
1192
1193 qdma_dev->state = 1;
1194
1195 return 0;
1196 }
1197
1198 static int
check_devargs_handler(__rte_unused const char * key,const char * value,__rte_unused void * opaque)1199 check_devargs_handler(__rte_unused const char *key, const char *value,
1200 __rte_unused void *opaque)
1201 {
1202 if (strcmp(value, "1"))
1203 return -1;
1204
1205 return 0;
1206 }
1207
1208 static int
dpaa2_get_devargs(struct rte_devargs * devargs,const char * key)1209 dpaa2_get_devargs(struct rte_devargs *devargs, const char *key)
1210 {
1211 struct rte_kvargs *kvlist;
1212
1213 if (!devargs)
1214 return 0;
1215
1216 kvlist = rte_kvargs_parse(devargs->args, NULL);
1217 if (!kvlist)
1218 return 0;
1219
1220 if (!rte_kvargs_count(kvlist, key)) {
1221 rte_kvargs_free(kvlist);
1222 return 0;
1223 }
1224
1225 if (rte_kvargs_process(kvlist, key,
1226 check_devargs_handler, NULL) < 0) {
1227 rte_kvargs_free(kvlist);
1228 return 0;
1229 }
1230 rte_kvargs_free(kvlist);
1231
1232 return 1;
1233 }
1234
1235 static int
dpaa2_qdma_queue_setup(struct rte_rawdev * rawdev,__rte_unused uint16_t queue_id,rte_rawdev_obj_t queue_conf,size_t conf_size)1236 dpaa2_qdma_queue_setup(struct rte_rawdev *rawdev,
1237 __rte_unused uint16_t queue_id,
1238 rte_rawdev_obj_t queue_conf,
1239 size_t conf_size)
1240 {
1241 char ring_name[32];
1242 char pool_name[64];
1243 int i;
1244 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1245 struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
1246 struct rte_qdma_queue_config *q_config =
1247 (struct rte_qdma_queue_config *)queue_conf;
1248 uint32_t pool_size;
1249
1250 DPAA2_QDMA_FUNC_TRACE();
1251
1252 if (conf_size != sizeof(*q_config)) {
1253 DPAA2_QDMA_ERR("Config size mismatch. Expected %" PRIu64
1254 ", Got: %" PRIu64, (uint64_t)(sizeof(*q_config)),
1255 (uint64_t)conf_size);
1256 return -EINVAL;
1257 }
1258
1259 rte_spinlock_lock(&qdma_dev->lock);
1260
1261 /* Get a free Virtual Queue */
1262 for (i = 0; i < qdma_dev->max_vqs; i++) {
1263 if (qdma_dev->vqs[i].in_use == 0)
1264 break;
1265 }
1266
1267 /* Return in case no VQ is free */
1268 if (i == qdma_dev->max_vqs) {
1269 rte_spinlock_unlock(&qdma_dev->lock);
1270 DPAA2_QDMA_ERR("Unable to get lock on QDMA device");
1271 return -ENODEV;
1272 }
1273
1274 if (q_config->flags & RTE_QDMA_VQ_FD_SG_FORMAT) {
1275 if (!(q_config->flags & RTE_QDMA_VQ_EXCLUSIVE_PQ)) {
1276 DPAA2_QDMA_ERR(
1277 "qDMA SG format only supports physical queue!");
1278 rte_spinlock_unlock(&qdma_dev->lock);
1279 return -ENODEV;
1280 }
1281 if (!(q_config->flags & RTE_QDMA_VQ_FD_LONG_FORMAT)) {
1282 DPAA2_QDMA_ERR(
1283 "qDMA SG format only supports long FD format!");
1284 rte_spinlock_unlock(&qdma_dev->lock);
1285 return -ENODEV;
1286 }
1287 pool_size = QDMA_FLE_SG_POOL_SIZE;
1288 } else {
1289 pool_size = QDMA_FLE_SINGLE_POOL_SIZE;
1290 }
1291
1292 if (q_config->flags & RTE_QDMA_VQ_EXCLUSIVE_PQ) {
1293 /* Allocate HW queue for a VQ */
1294 qdma_dev->vqs[i].hw_queue = alloc_hw_queue(q_config->lcore_id);
1295 qdma_dev->vqs[i].exclusive_hw_queue = 1;
1296 } else {
1297 /* Allocate a Ring for Virtual Queue in VQ mode */
1298 snprintf(ring_name, sizeof(ring_name), "status ring %d", i);
1299 qdma_dev->vqs[i].status_ring = rte_ring_create(ring_name,
1300 qdma_dev->fle_queue_pool_cnt, rte_socket_id(), 0);
1301 if (!qdma_dev->vqs[i].status_ring) {
1302 DPAA2_QDMA_ERR("Status ring creation failed for vq");
1303 rte_spinlock_unlock(&qdma_dev->lock);
1304 return rte_errno;
1305 }
1306
1307 /* Get a HW queue (shared) for a VQ */
1308 qdma_dev->vqs[i].hw_queue = get_hw_queue(qdma_dev,
1309 q_config->lcore_id);
1310 qdma_dev->vqs[i].exclusive_hw_queue = 0;
1311 }
1312
1313 if (qdma_dev->vqs[i].hw_queue == NULL) {
1314 DPAA2_QDMA_ERR("No H/W queue available for VQ");
1315 rte_ring_free(qdma_dev->vqs[i].status_ring);
1316 qdma_dev->vqs[i].status_ring = NULL;
1317 rte_spinlock_unlock(&qdma_dev->lock);
1318 return -ENODEV;
1319 }
1320
1321 snprintf(pool_name, sizeof(pool_name),
1322 "qdma_fle_pool%u_queue%d", getpid(), i);
1323 qdma_dev->vqs[i].fle_pool = rte_mempool_create(pool_name,
1324 qdma_dev->fle_queue_pool_cnt, pool_size,
1325 QDMA_FLE_CACHE_SIZE(qdma_dev->fle_queue_pool_cnt), 0,
1326 NULL, NULL, NULL, NULL, SOCKET_ID_ANY, 0);
1327 if (!qdma_dev->vqs[i].fle_pool) {
1328 DPAA2_QDMA_ERR("qdma_fle_pool create failed");
1329 rte_spinlock_unlock(&qdma_dev->lock);
1330 return -ENOMEM;
1331 }
1332
1333 qdma_dev->vqs[i].flags = q_config->flags;
1334 qdma_dev->vqs[i].in_use = 1;
1335 qdma_dev->vqs[i].lcore_id = q_config->lcore_id;
1336 memset(&qdma_dev->vqs[i].rbp, 0, sizeof(struct rte_qdma_rbp));
1337
1338 if (q_config->flags & RTE_QDMA_VQ_FD_LONG_FORMAT) {
1339 if (q_config->flags & RTE_QDMA_VQ_FD_SG_FORMAT) {
1340 qdma_dev->vqs[i].set_fd = dpdmai_dev_set_sg_fd_lf;
1341 qdma_dev->vqs[i].get_job = dpdmai_dev_get_sg_job_lf;
1342 } else {
1343 if (q_config->flags & RTE_QDMA_VQ_NO_RESPONSE)
1344 qdma_dev->vqs[i].set_fd =
1345 dpdmai_dev_set_multi_fd_lf_no_rsp;
1346 else
1347 qdma_dev->vqs[i].set_fd =
1348 dpdmai_dev_set_multi_fd_lf;
1349 qdma_dev->vqs[i].get_job = dpdmai_dev_get_single_job_lf;
1350 }
1351 } else {
1352 qdma_dev->vqs[i].set_fd = dpdmai_dev_set_fd_us;
1353 qdma_dev->vqs[i].get_job = dpdmai_dev_get_job_us;
1354 }
1355 if (dpaa2_get_devargs(rawdev->device->devargs,
1356 DPAA2_QDMA_NO_PREFETCH) ||
1357 (getenv("DPAA2_NO_QDMA_PREFETCH_RX"))) {
1358 /* If no prefetch is configured. */
1359 qdma_dev->vqs[i].dequeue_job =
1360 dpdmai_dev_dequeue_multijob_no_prefetch;
1361 DPAA2_QDMA_INFO("No Prefetch RX Mode enabled");
1362 } else {
1363 qdma_dev->vqs[i].dequeue_job =
1364 dpdmai_dev_dequeue_multijob_prefetch;
1365 }
1366
1367 qdma_dev->vqs[i].enqueue_job = dpdmai_dev_enqueue_multi;
1368
1369 if (q_config->rbp != NULL)
1370 memcpy(&qdma_dev->vqs[i].rbp, q_config->rbp,
1371 sizeof(struct rte_qdma_rbp));
1372
1373 rte_spinlock_unlock(&qdma_dev->lock);
1374
1375 return i;
1376 }
1377
1378 static int
dpaa2_qdma_enqueue(struct rte_rawdev * rawdev,__rte_unused struct rte_rawdev_buf ** buffers,unsigned int nb_jobs,rte_rawdev_obj_t context)1379 dpaa2_qdma_enqueue(struct rte_rawdev *rawdev,
1380 __rte_unused struct rte_rawdev_buf **buffers,
1381 unsigned int nb_jobs,
1382 rte_rawdev_obj_t context)
1383 {
1384 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1385 struct rte_qdma_enqdeq *e_context =
1386 (struct rte_qdma_enqdeq *)context;
1387 struct qdma_virt_queue *qdma_vq =
1388 &dpdmai_dev->qdma_dev->vqs[e_context->vq_id];
1389 int ret;
1390
1391 ret = qdma_vq->enqueue_job(qdma_vq, e_context->job, nb_jobs);
1392 if (ret < 0) {
1393 DPAA2_QDMA_ERR("DPDMAI device enqueue failed: %d", ret);
1394 return ret;
1395 }
1396
1397 qdma_vq->num_enqueues += ret;
1398
1399 return ret;
1400 }
1401
1402 static int
dpaa2_qdma_dequeue(struct rte_rawdev * rawdev,__rte_unused struct rte_rawdev_buf ** buffers,unsigned int nb_jobs,rte_rawdev_obj_t cntxt)1403 dpaa2_qdma_dequeue(struct rte_rawdev *rawdev,
1404 __rte_unused struct rte_rawdev_buf **buffers,
1405 unsigned int nb_jobs,
1406 rte_rawdev_obj_t cntxt)
1407 {
1408 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1409 struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
1410 struct rte_qdma_enqdeq *context =
1411 (struct rte_qdma_enqdeq *)cntxt;
1412 struct qdma_virt_queue *qdma_vq = &qdma_dev->vqs[context->vq_id];
1413 struct qdma_virt_queue *temp_qdma_vq;
1414 int ret = 0, i;
1415 unsigned int ring_count;
1416
1417 if (qdma_vq->flags & RTE_QDMA_VQ_FD_SG_FORMAT) {
1418 /** Make sure there are enough space to get jobs.*/
1419 if (unlikely(nb_jobs < DPAA2_QDMA_MAX_SG_NB))
1420 return -EINVAL;
1421 }
1422
1423 /* Only dequeue when there are pending jobs on VQ */
1424 if (qdma_vq->num_enqueues == qdma_vq->num_dequeues)
1425 return 0;
1426
1427 if (!(qdma_vq->flags & RTE_QDMA_VQ_FD_SG_FORMAT) &&
1428 qdma_vq->num_enqueues < (qdma_vq->num_dequeues + nb_jobs))
1429 nb_jobs = (qdma_vq->num_enqueues - qdma_vq->num_dequeues);
1430
1431 if (qdma_vq->exclusive_hw_queue) {
1432 /* In case of exclusive queue directly fetch from HW queue */
1433 ret = qdma_vq->dequeue_job(qdma_vq, NULL,
1434 context->job, nb_jobs);
1435 if (ret < 0) {
1436 DPAA2_QDMA_ERR(
1437 "Dequeue from DPDMAI device failed: %d", ret);
1438 return ret;
1439 }
1440 qdma_vq->num_dequeues += ret;
1441 } else {
1442 uint16_t temp_vq_id[RTE_QDMA_BURST_NB_MAX];
1443 /*
1444 * Get the QDMA completed jobs from the software ring.
1445 * In case they are not available on the ring poke the HW
1446 * to fetch completed jobs from corresponding HW queues
1447 */
1448 ring_count = rte_ring_count(qdma_vq->status_ring);
1449 if (ring_count < nb_jobs) {
1450 /* TODO - How to have right budget */
1451 ret = qdma_vq->dequeue_job(qdma_vq,
1452 temp_vq_id, context->job, nb_jobs);
1453 for (i = 0; i < ret; i++) {
1454 temp_qdma_vq = &qdma_dev->vqs[temp_vq_id[i]];
1455 rte_ring_enqueue(temp_qdma_vq->status_ring,
1456 (void *)(context->job[i]));
1457 }
1458 ring_count = rte_ring_count(
1459 qdma_vq->status_ring);
1460 }
1461
1462 if (ring_count) {
1463 /* Dequeue job from the software ring
1464 * to provide to the user
1465 */
1466 ret = rte_ring_dequeue_bulk(qdma_vq->status_ring,
1467 (void **)context->job,
1468 ring_count, NULL);
1469 if (ret)
1470 qdma_vq->num_dequeues += ret;
1471 }
1472 }
1473
1474 return ret;
1475 }
1476
1477 void
rte_qdma_vq_stats(struct rte_rawdev * rawdev,uint16_t vq_id,struct rte_qdma_vq_stats * vq_status)1478 rte_qdma_vq_stats(struct rte_rawdev *rawdev,
1479 uint16_t vq_id,
1480 struct rte_qdma_vq_stats *vq_status)
1481 {
1482 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1483 struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
1484 struct qdma_virt_queue *qdma_vq = &qdma_dev->vqs[vq_id];
1485
1486 if (qdma_vq->in_use) {
1487 vq_status->exclusive_hw_queue = qdma_vq->exclusive_hw_queue;
1488 vq_status->lcore_id = qdma_vq->lcore_id;
1489 vq_status->num_enqueues = qdma_vq->num_enqueues;
1490 vq_status->num_dequeues = qdma_vq->num_dequeues;
1491 vq_status->num_pending_jobs = vq_status->num_enqueues -
1492 vq_status->num_dequeues;
1493 }
1494 }
1495
1496 static int
dpaa2_qdma_queue_release(struct rte_rawdev * rawdev,uint16_t vq_id)1497 dpaa2_qdma_queue_release(struct rte_rawdev *rawdev,
1498 uint16_t vq_id)
1499 {
1500 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1501 struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
1502
1503 struct qdma_virt_queue *qdma_vq = &qdma_dev->vqs[vq_id];
1504
1505 DPAA2_QDMA_FUNC_TRACE();
1506
1507 /* In case there are pending jobs on any VQ, return -EBUSY */
1508 if (qdma_vq->num_enqueues != qdma_vq->num_dequeues)
1509 return -EBUSY;
1510
1511 rte_spinlock_lock(&qdma_dev->lock);
1512
1513 if (qdma_vq->exclusive_hw_queue)
1514 free_hw_queue(qdma_vq->hw_queue);
1515 else {
1516 rte_ring_free(qdma_vq->status_ring);
1517
1518 put_hw_queue(qdma_vq->hw_queue);
1519 }
1520
1521 rte_mempool_free(qdma_vq->fle_pool);
1522
1523 memset(qdma_vq, 0, sizeof(struct qdma_virt_queue));
1524
1525 rte_spinlock_unlock(&qdma_dev->lock);
1526
1527 return 0;
1528 }
1529
1530 static void
dpaa2_qdma_stop(struct rte_rawdev * rawdev)1531 dpaa2_qdma_stop(struct rte_rawdev *rawdev)
1532 {
1533 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1534 struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
1535
1536 DPAA2_QDMA_FUNC_TRACE();
1537
1538 qdma_dev->state = 0;
1539 }
1540
1541 static int
dpaa2_qdma_close(struct rte_rawdev * rawdev)1542 dpaa2_qdma_close(struct rte_rawdev *rawdev)
1543 {
1544 DPAA2_QDMA_FUNC_TRACE();
1545
1546 dpaa2_qdma_reset(rawdev);
1547
1548 return 0;
1549 }
1550
1551 static struct rte_rawdev_ops dpaa2_qdma_ops = {
1552 .dev_configure = dpaa2_qdma_configure,
1553 .dev_start = dpaa2_qdma_start,
1554 .dev_stop = dpaa2_qdma_stop,
1555 .dev_reset = dpaa2_qdma_reset,
1556 .dev_close = dpaa2_qdma_close,
1557 .queue_setup = dpaa2_qdma_queue_setup,
1558 .queue_release = dpaa2_qdma_queue_release,
1559 .attr_get = dpaa2_qdma_attr_get,
1560 .enqueue_bufs = dpaa2_qdma_enqueue,
1561 .dequeue_bufs = dpaa2_qdma_dequeue,
1562 };
1563
1564 static int
add_hw_queues_to_list(struct dpaa2_dpdmai_dev * dpdmai_dev)1565 add_hw_queues_to_list(struct dpaa2_dpdmai_dev *dpdmai_dev)
1566 {
1567 struct qdma_hw_queue *queue;
1568 int i;
1569
1570 DPAA2_QDMA_FUNC_TRACE();
1571
1572 for (i = 0; i < dpdmai_dev->num_queues; i++) {
1573 queue = rte_zmalloc(NULL, sizeof(struct qdma_hw_queue), 0);
1574 if (!queue) {
1575 DPAA2_QDMA_ERR(
1576 "Memory allocation failed for QDMA queue");
1577 return -ENOMEM;
1578 }
1579
1580 queue->dpdmai_dev = dpdmai_dev;
1581 queue->queue_id = i;
1582
1583 TAILQ_INSERT_TAIL(&qdma_queue_list, queue, next);
1584 dpdmai_dev->qdma_dev->num_hw_queues++;
1585 }
1586
1587 return 0;
1588 }
1589
1590 static void
remove_hw_queues_from_list(struct dpaa2_dpdmai_dev * dpdmai_dev)1591 remove_hw_queues_from_list(struct dpaa2_dpdmai_dev *dpdmai_dev)
1592 {
1593 struct qdma_hw_queue *queue = NULL;
1594 struct qdma_hw_queue *tqueue = NULL;
1595
1596 DPAA2_QDMA_FUNC_TRACE();
1597
1598 RTE_TAILQ_FOREACH_SAFE(queue, &qdma_queue_list, next, tqueue) {
1599 if (queue->dpdmai_dev == dpdmai_dev) {
1600 TAILQ_REMOVE(&qdma_queue_list, queue, next);
1601 rte_free(queue);
1602 queue = NULL;
1603 }
1604 }
1605 }
1606
1607 static int
dpaa2_dpdmai_dev_uninit(struct rte_rawdev * rawdev)1608 dpaa2_dpdmai_dev_uninit(struct rte_rawdev *rawdev)
1609 {
1610 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1611 int ret, i;
1612
1613 DPAA2_QDMA_FUNC_TRACE();
1614
1615 /* Remove HW queues from global list */
1616 remove_hw_queues_from_list(dpdmai_dev);
1617
1618 ret = dpdmai_disable(&dpdmai_dev->dpdmai, CMD_PRI_LOW,
1619 dpdmai_dev->token);
1620 if (ret)
1621 DPAA2_QDMA_ERR("dmdmai disable failed");
1622
1623 /* Set up the DQRR storage for Rx */
1624 for (i = 0; i < dpdmai_dev->num_queues; i++) {
1625 struct dpaa2_queue *rxq = &(dpdmai_dev->rx_queue[i]);
1626
1627 if (rxq->q_storage) {
1628 dpaa2_free_dq_storage(rxq->q_storage);
1629 rte_free(rxq->q_storage);
1630 }
1631 }
1632
1633 /* Close the device at underlying layer*/
1634 ret = dpdmai_close(&dpdmai_dev->dpdmai, CMD_PRI_LOW, dpdmai_dev->token);
1635 if (ret)
1636 DPAA2_QDMA_ERR("Failure closing dpdmai device");
1637
1638 return 0;
1639 }
1640
1641 static int
dpaa2_dpdmai_dev_init(struct rte_rawdev * rawdev,int dpdmai_id)1642 dpaa2_dpdmai_dev_init(struct rte_rawdev *rawdev, int dpdmai_id)
1643 {
1644 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1645 struct dpdmai_rx_queue_cfg rx_queue_cfg;
1646 struct dpdmai_attr attr;
1647 struct dpdmai_rx_queue_attr rx_attr;
1648 struct dpdmai_tx_queue_attr tx_attr;
1649 int ret, i;
1650
1651 DPAA2_QDMA_FUNC_TRACE();
1652
1653 /* Open DPDMAI device */
1654 dpdmai_dev->dpdmai_id = dpdmai_id;
1655 dpdmai_dev->dpdmai.regs = dpaa2_get_mcp_ptr(MC_PORTAL_INDEX);
1656 dpdmai_dev->qdma_dev = &q_dev;
1657 ret = dpdmai_open(&dpdmai_dev->dpdmai, CMD_PRI_LOW,
1658 dpdmai_dev->dpdmai_id, &dpdmai_dev->token);
1659 if (ret) {
1660 DPAA2_QDMA_ERR("dpdmai_open() failed with err: %d", ret);
1661 return ret;
1662 }
1663
1664 /* Get DPDMAI attributes */
1665 ret = dpdmai_get_attributes(&dpdmai_dev->dpdmai, CMD_PRI_LOW,
1666 dpdmai_dev->token, &attr);
1667 if (ret) {
1668 DPAA2_QDMA_ERR("dpdmai get attributes failed with err: %d",
1669 ret);
1670 goto init_err;
1671 }
1672 dpdmai_dev->num_queues = attr.num_of_queues;
1673
1674 /* Set up Rx Queues */
1675 for (i = 0; i < dpdmai_dev->num_queues; i++) {
1676 struct dpaa2_queue *rxq;
1677
1678 memset(&rx_queue_cfg, 0, sizeof(struct dpdmai_rx_queue_cfg));
1679 ret = dpdmai_set_rx_queue(&dpdmai_dev->dpdmai,
1680 CMD_PRI_LOW,
1681 dpdmai_dev->token,
1682 i, 0, &rx_queue_cfg);
1683 if (ret) {
1684 DPAA2_QDMA_ERR("Setting Rx queue failed with err: %d",
1685 ret);
1686 goto init_err;
1687 }
1688
1689 /* Allocate DQ storage for the DPDMAI Rx queues */
1690 rxq = &(dpdmai_dev->rx_queue[i]);
1691 rxq->q_storage = rte_malloc("dq_storage",
1692 sizeof(struct queue_storage_info_t),
1693 RTE_CACHE_LINE_SIZE);
1694 if (!rxq->q_storage) {
1695 DPAA2_QDMA_ERR("q_storage allocation failed");
1696 ret = -ENOMEM;
1697 goto init_err;
1698 }
1699
1700 memset(rxq->q_storage, 0, sizeof(struct queue_storage_info_t));
1701 ret = dpaa2_alloc_dq_storage(rxq->q_storage);
1702 if (ret) {
1703 DPAA2_QDMA_ERR("dpaa2_alloc_dq_storage failed");
1704 goto init_err;
1705 }
1706 }
1707
1708 /* Get Rx and Tx queues FQID's */
1709 for (i = 0; i < dpdmai_dev->num_queues; i++) {
1710 ret = dpdmai_get_rx_queue(&dpdmai_dev->dpdmai, CMD_PRI_LOW,
1711 dpdmai_dev->token, i, 0, &rx_attr);
1712 if (ret) {
1713 DPAA2_QDMA_ERR("Reading device failed with err: %d",
1714 ret);
1715 goto init_err;
1716 }
1717 dpdmai_dev->rx_queue[i].fqid = rx_attr.fqid;
1718
1719 ret = dpdmai_get_tx_queue(&dpdmai_dev->dpdmai, CMD_PRI_LOW,
1720 dpdmai_dev->token, i, 0, &tx_attr);
1721 if (ret) {
1722 DPAA2_QDMA_ERR("Reading device failed with err: %d",
1723 ret);
1724 goto init_err;
1725 }
1726 dpdmai_dev->tx_queue[i].fqid = tx_attr.fqid;
1727 }
1728
1729 /* Enable the device */
1730 ret = dpdmai_enable(&dpdmai_dev->dpdmai, CMD_PRI_LOW,
1731 dpdmai_dev->token);
1732 if (ret) {
1733 DPAA2_QDMA_ERR("Enabling device failed with err: %d", ret);
1734 goto init_err;
1735 }
1736
1737 /* Add the HW queue to the global list */
1738 ret = add_hw_queues_to_list(dpdmai_dev);
1739 if (ret) {
1740 DPAA2_QDMA_ERR("Adding H/W queue to list failed");
1741 goto init_err;
1742 }
1743
1744 if (!dpaa2_coherent_no_alloc_cache) {
1745 if (dpaa2_svr_family == SVR_LX2160A) {
1746 dpaa2_coherent_no_alloc_cache =
1747 DPAA2_LX2_COHERENT_NO_ALLOCATE_CACHE;
1748 dpaa2_coherent_alloc_cache =
1749 DPAA2_LX2_COHERENT_ALLOCATE_CACHE;
1750 } else {
1751 dpaa2_coherent_no_alloc_cache =
1752 DPAA2_COHERENT_NO_ALLOCATE_CACHE;
1753 dpaa2_coherent_alloc_cache =
1754 DPAA2_COHERENT_ALLOCATE_CACHE;
1755 }
1756 }
1757
1758 DPAA2_QDMA_DEBUG("Initialized dpdmai object successfully");
1759
1760 rte_spinlock_init(&dpdmai_dev->qdma_dev->lock);
1761
1762 return 0;
1763 init_err:
1764 dpaa2_dpdmai_dev_uninit(rawdev);
1765 return ret;
1766 }
1767
1768 static int
rte_dpaa2_qdma_probe(struct rte_dpaa2_driver * dpaa2_drv,struct rte_dpaa2_device * dpaa2_dev)1769 rte_dpaa2_qdma_probe(struct rte_dpaa2_driver *dpaa2_drv,
1770 struct rte_dpaa2_device *dpaa2_dev)
1771 {
1772 struct rte_rawdev *rawdev;
1773 int ret;
1774
1775 DPAA2_QDMA_FUNC_TRACE();
1776
1777 rawdev = rte_rawdev_pmd_allocate(dpaa2_dev->device.name,
1778 sizeof(struct dpaa2_dpdmai_dev),
1779 rte_socket_id());
1780 if (!rawdev) {
1781 DPAA2_QDMA_ERR("Unable to allocate rawdevice");
1782 return -EINVAL;
1783 }
1784
1785 dpaa2_dev->rawdev = rawdev;
1786 rawdev->dev_ops = &dpaa2_qdma_ops;
1787 rawdev->device = &dpaa2_dev->device;
1788 rawdev->driver_name = dpaa2_drv->driver.name;
1789
1790 /* Invoke PMD device initialization function */
1791 ret = dpaa2_dpdmai_dev_init(rawdev, dpaa2_dev->object_id);
1792 if (ret) {
1793 rte_rawdev_pmd_release(rawdev);
1794 return ret;
1795 }
1796
1797 /* Reset the QDMA device */
1798 ret = dpaa2_qdma_reset(rawdev);
1799 if (ret) {
1800 DPAA2_QDMA_ERR("Resetting QDMA failed");
1801 return ret;
1802 }
1803
1804 return 0;
1805 }
1806
1807 static int
rte_dpaa2_qdma_remove(struct rte_dpaa2_device * dpaa2_dev)1808 rte_dpaa2_qdma_remove(struct rte_dpaa2_device *dpaa2_dev)
1809 {
1810 struct rte_rawdev *rawdev = dpaa2_dev->rawdev;
1811 int ret;
1812
1813 DPAA2_QDMA_FUNC_TRACE();
1814
1815 dpaa2_dpdmai_dev_uninit(rawdev);
1816
1817 ret = rte_rawdev_pmd_release(rawdev);
1818 if (ret)
1819 DPAA2_QDMA_ERR("Device cleanup failed");
1820
1821 return 0;
1822 }
1823
1824 static struct rte_dpaa2_driver rte_dpaa2_qdma_pmd = {
1825 .drv_flags = RTE_DPAA2_DRV_IOVA_AS_VA,
1826 .drv_type = DPAA2_QDMA,
1827 .probe = rte_dpaa2_qdma_probe,
1828 .remove = rte_dpaa2_qdma_remove,
1829 };
1830
1831 RTE_PMD_REGISTER_DPAA2(dpaa2_qdma, rte_dpaa2_qdma_pmd);
1832 RTE_PMD_REGISTER_PARAM_STRING(dpaa2_qdma,
1833 "no_prefetch=<int> ");
1834 RTE_LOG_REGISTER(dpaa2_qdma_logtype, pmd.raw.dpaa2.qdma, INFO);
1835