1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2019 Intel Corporation
7 #include <rte_common.h>
10 #include <rte_malloc.h>
11 #include <rte_mempool.h>
12 #include <rte_errno.h>
14 #include <rte_bus_pci.h>
15 #include <rte_byteorder.h>
16 #ifdef RTE_BBDEV_OFFLOAD_COST
17 #include <rte_cycles.h>
20 #include <rte_bbdev.h>
21 #include <rte_bbdev_pmd.h>
23 #include "fpga_lte_fec.h"
25 #ifdef RTE_LIBRTE_BBDEV_DEBUG
26 RTE_LOG_REGISTER(fpga_lte_fec_logtype, pmd.bb.fpga_lte_fec, DEBUG);
28 RTE_LOG_REGISTER(fpga_lte_fec_logtype, pmd.bb.fpga_lte_fec, NOTICE);
31 /* Helper macro for logging */
32 #define rte_bbdev_log(level, fmt, ...) \
33 rte_log(RTE_LOG_ ## level, fpga_lte_fec_logtype, fmt "\n", \
36 #ifdef RTE_LIBRTE_BBDEV_DEBUG
37 #define rte_bbdev_log_debug(fmt, ...) \
38 rte_bbdev_log(DEBUG, "fpga_lte_fec: " fmt, \
41 #define rte_bbdev_log_debug(fmt, ...)
44 /* FPGA LTE FEC driver names */
45 #define FPGA_LTE_FEC_PF_DRIVER_NAME intel_fpga_lte_fec_pf
46 #define FPGA_LTE_FEC_VF_DRIVER_NAME intel_fpga_lte_fec_vf
48 /* FPGA LTE FEC PCI vendor & device IDs */
49 #define FPGA_LTE_FEC_VENDOR_ID (0x1172)
50 #define FPGA_LTE_FEC_PF_DEVICE_ID (0x5052)
51 #define FPGA_LTE_FEC_VF_DEVICE_ID (0x5050)
53 /* Align DMA descriptors to 256 bytes - cache-aligned */
54 #define FPGA_RING_DESC_ENTRY_LENGTH (8)
55 /* Ring size is in 256 bits (32 bytes) units */
56 #define FPGA_RING_DESC_LEN_UNIT_BYTES (32)
57 /* Maximum size of queue */
58 #define FPGA_RING_MAX_SIZE (1024)
59 #define FPGA_FLR_TIMEOUT_UNIT (16.384)
61 #define FPGA_NUM_UL_QUEUES (32)
62 #define FPGA_NUM_DL_QUEUES (32)
63 #define FPGA_TOTAL_NUM_QUEUES (FPGA_NUM_UL_QUEUES + FPGA_NUM_DL_QUEUES)
64 #define FPGA_NUM_INTR_VEC (FPGA_TOTAL_NUM_QUEUES - RTE_INTR_VEC_RXTX_OFFSET)
66 #define FPGA_INVALID_HW_QUEUE_ID (0xFFFFFFFF)
68 #define FPGA_QUEUE_FLUSH_TIMEOUT_US (1000)
69 #define FPGA_TIMEOUT_CHECK_INTERVAL (5)
71 /* FPGA LTE FEC Register mapping on BAR0 */
73 FPGA_LTE_FEC_VERSION_ID = 0x00000000, /* len: 4B */
74 FPGA_LTE_FEC_CONFIGURATION = 0x00000004, /* len: 2B */
75 FPGA_LTE_FEC_QUEUE_PF_VF_MAP_DONE = 0x00000008, /* len: 1B */
76 FPGA_LTE_FEC_LOAD_BALANCE_FACTOR = 0x0000000a, /* len: 2B */
77 FPGA_LTE_FEC_RING_DESC_LEN = 0x0000000c, /* len: 2B */
78 FPGA_LTE_FEC_FLR_TIME_OUT = 0x0000000e, /* len: 2B */
79 FPGA_LTE_FEC_VFQ_FLUSH_STATUS_LW = 0x00000018, /* len: 4B */
80 FPGA_LTE_FEC_VFQ_FLUSH_STATUS_HI = 0x0000001c, /* len: 4B */
81 FPGA_LTE_FEC_VF0_DEBUG = 0x00000020, /* len: 4B */
82 FPGA_LTE_FEC_VF1_DEBUG = 0x00000024, /* len: 4B */
83 FPGA_LTE_FEC_VF2_DEBUG = 0x00000028, /* len: 4B */
84 FPGA_LTE_FEC_VF3_DEBUG = 0x0000002c, /* len: 4B */
85 FPGA_LTE_FEC_VF4_DEBUG = 0x00000030, /* len: 4B */
86 FPGA_LTE_FEC_VF5_DEBUG = 0x00000034, /* len: 4B */
87 FPGA_LTE_FEC_VF6_DEBUG = 0x00000038, /* len: 4B */
88 FPGA_LTE_FEC_VF7_DEBUG = 0x0000003c, /* len: 4B */
89 FPGA_LTE_FEC_QUEUE_MAP = 0x00000040, /* len: 256B */
90 FPGA_LTE_FEC_RING_CTRL_REGS = 0x00000200 /* len: 2048B */
93 /* FPGA LTE FEC Ring Control Registers */
95 FPGA_LTE_FEC_RING_HEAD_ADDR = 0x00000008,
96 FPGA_LTE_FEC_RING_SIZE = 0x00000010,
97 FPGA_LTE_FEC_RING_MISC = 0x00000014,
98 FPGA_LTE_FEC_RING_ENABLE = 0x00000015,
99 FPGA_LTE_FEC_RING_FLUSH_QUEUE_EN = 0x00000016,
100 FPGA_LTE_FEC_RING_SHADOW_TAIL = 0x00000018,
101 FPGA_LTE_FEC_RING_HEAD_POINT = 0x0000001C
104 /* FPGA LTE FEC DESCRIPTOR ERROR */
106 DESC_ERR_NO_ERR = 0x0,
107 DESC_ERR_K_OUT_OF_RANGE = 0x1,
108 DESC_ERR_K_NOT_NORMAL = 0x2,
109 DESC_ERR_KPAI_NOT_NORMAL = 0x3,
110 DESC_ERR_DESC_OFFSET_ERR = 0x4,
111 DESC_ERR_DESC_READ_FAIL = 0x8,
112 DESC_ERR_DESC_READ_TIMEOUT = 0x9,
113 DESC_ERR_DESC_READ_TLP_POISONED = 0xA,
114 DESC_ERR_CB_READ_FAIL = 0xC,
115 DESC_ERR_CB_READ_TIMEOUT = 0xD,
116 DESC_ERR_CB_READ_TLP_POISONED = 0xE
119 /* FPGA LTE FEC DMA Encoding Request Descriptor */
120 struct __rte_packed fpga_dma_enc_desc {
128 uint32_t bypass_rm:1,
136 uint32_t out_addr_lw;
137 uint32_t out_addr_hi;
143 /* Virtual addresses used to retrieve SW context info */
145 /* Stores information about total number of Code Blocks
146 * in currently processed Transport Block
151 uint8_t sw_ctxt[FPGA_RING_DESC_LEN_UNIT_BYTES *
152 (FPGA_RING_DESC_ENTRY_LENGTH - 1)];
156 /* FPGA LTE FEC DMA Decoding Request Descriptor */
157 struct __rte_packed fpga_dma_dec_desc {
169 uint32_t bypass_rm:1,
177 uint32_t out_addr_lw;
178 uint32_t out_addr_hi;
184 /* Virtual addresses used to retrieve SW context info */
186 /* Stores information about total number of Code Blocks
187 * in currently processed Transport Block
192 uint32_t sw_ctxt[8 * (FPGA_RING_DESC_ENTRY_LENGTH - 1)];
196 /* FPGA LTE DMA Descriptor */
197 union fpga_dma_desc {
198 struct fpga_dma_enc_desc enc_req;
199 struct fpga_dma_dec_desc dec_req;
202 /* FPGA LTE FEC Ring Control Register */
203 struct __rte_packed fpga_ring_ctrl_reg {
204 uint64_t ring_base_addr;
205 uint64_t ring_head_addr;
206 uint16_t ring_size:11;
208 union { /* Miscellaneous register */
210 uint8_t max_ul_dec:5,
215 uint8_t flush_queue_en;
217 uint16_t shadow_tail;
224 /* Private data structure for each FPGA FEC device */
225 struct fpga_lte_fec_device {
226 /** Base address of MMIO registers (BAR0) */
228 /** Base address of memory for sw rings */
230 /** Physical address of sw_rings */
231 rte_iova_t sw_rings_phys;
232 /** Number of bytes available for each queue in device. */
233 uint32_t sw_ring_size;
234 /** Max number of entries available for each queue in device */
235 uint32_t sw_ring_max_depth;
236 /** Base address of response tail pointer buffer */
238 /** Physical address of tail pointers */
239 rte_iova_t tail_ptr_phys;
240 /** Queues flush completion flag */
241 uint64_t *flush_queue_status;
242 /* Bitmap capturing which Queues are bound to the PF/VF */
243 uint64_t q_bound_bit_map;
244 /* Bitmap capturing which Queues have already been assigned */
245 uint64_t q_assigned_bit_map;
246 /** True if this is a PF FPGA FEC device */
250 /* Structure associated with each queue. */
251 struct __rte_cache_aligned fpga_queue {
252 struct fpga_ring_ctrl_reg ring_ctrl_reg; /* Ring Control Register */
253 union fpga_dma_desc *ring_addr; /* Virtual address of software ring */
254 uint64_t *ring_head_addr; /* Virtual address of completion_head */
255 uint64_t shadow_completion_head; /* Shadow completion head value */
256 uint16_t head_free_desc; /* Ring head */
257 uint16_t tail; /* Ring tail */
258 /* Mask used to wrap enqueued descriptors on the sw ring */
259 uint32_t sw_ring_wrap_mask;
260 uint32_t irq_enable; /* Enable ops dequeue interrupts if set to 1 */
261 uint8_t q_idx; /* Queue index */
262 struct fpga_lte_fec_device *d;
263 /* MMIO register of shadow_tail used to enqueue descriptors */
264 void *shadow_tail_addr;
267 /* Write to 16 bit MMIO register address */
269 mmio_write_16(void *addr, uint16_t value)
271 *((volatile uint16_t *)(addr)) = rte_cpu_to_le_16(value);
274 /* Write to 32 bit MMIO register address */
276 mmio_write_32(void *addr, uint32_t value)
278 *((volatile uint32_t *)(addr)) = rte_cpu_to_le_32(value);
281 /* Write to 64 bit MMIO register address */
283 mmio_write_64(void *addr, uint64_t value)
285 *((volatile uint64_t *)(addr)) = rte_cpu_to_le_64(value);
288 /* Write a 8 bit register of a FPGA LTE FEC device */
290 fpga_reg_write_8(void *mmio_base, uint32_t offset, uint8_t payload)
292 void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
293 *((volatile uint8_t *)(reg_addr)) = payload;
296 /* Write a 16 bit register of a FPGA LTE FEC device */
298 fpga_reg_write_16(void *mmio_base, uint32_t offset, uint16_t payload)
300 void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
301 mmio_write_16(reg_addr, payload);
304 /* Write a 32 bit register of a FPGA LTE FEC device */
306 fpga_reg_write_32(void *mmio_base, uint32_t offset, uint32_t payload)
308 void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
309 mmio_write_32(reg_addr, payload);
312 /* Write a 64 bit register of a FPGA LTE FEC device */
314 fpga_reg_write_64(void *mmio_base, uint32_t offset, uint64_t payload)
316 void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
317 mmio_write_64(reg_addr, payload);
320 /* Write a ring control register of a FPGA LTE FEC device */
322 fpga_ring_reg_write(void *mmio_base, uint32_t offset,
323 struct fpga_ring_ctrl_reg payload)
325 fpga_reg_write_64(mmio_base, offset, payload.ring_base_addr);
326 fpga_reg_write_64(mmio_base, offset + FPGA_LTE_FEC_RING_HEAD_ADDR,
327 payload.ring_head_addr);
328 fpga_reg_write_16(mmio_base, offset + FPGA_LTE_FEC_RING_SIZE,
330 fpga_reg_write_16(mmio_base, offset + FPGA_LTE_FEC_RING_HEAD_POINT,
332 fpga_reg_write_8(mmio_base, offset + FPGA_LTE_FEC_RING_FLUSH_QUEUE_EN,
333 payload.flush_queue_en);
334 fpga_reg_write_16(mmio_base, offset + FPGA_LTE_FEC_RING_SHADOW_TAIL,
335 payload.shadow_tail);
336 fpga_reg_write_8(mmio_base, offset + FPGA_LTE_FEC_RING_MISC,
338 fpga_reg_write_8(mmio_base, offset + FPGA_LTE_FEC_RING_ENABLE,
342 /* Read a register of FPGA LTE FEC device */
344 fpga_reg_read_32(void *mmio_base, uint32_t offset)
346 void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
347 uint32_t ret = *((volatile uint32_t *)(reg_addr));
348 return rte_le_to_cpu_32(ret);
351 #ifdef RTE_LIBRTE_BBDEV_DEBUG
352 /* Read a register of FPGA LTE FEC device */
354 fpga_reg_read_8(void *mmio_base, uint32_t offset)
356 void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
357 return *((volatile uint8_t *)(reg_addr));
360 /* Read a register of FPGA LTE FEC device */
362 fpga_reg_read_16(void *mmio_base, uint32_t offset)
364 void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
365 uint16_t ret = *((volatile uint16_t *)(reg_addr));
366 return rte_le_to_cpu_16(ret);
369 /* Read a register of FPGA LTE FEC device */
371 fpga_reg_read_64(void *mmio_base, uint32_t offset)
373 void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
374 uint64_t ret = *((volatile uint64_t *)(reg_addr));
375 return rte_le_to_cpu_64(ret);
378 /* Read Ring Control Register of FPGA LTE FEC device */
380 print_ring_reg_debug_info(void *mmio_base, uint32_t offset)
383 "FPGA MMIO base address @ %p | Ring Control Register @ offset = 0x%08"
384 PRIx32, mmio_base, offset);
386 "RING_BASE_ADDR = 0x%016"PRIx64,
387 fpga_reg_read_64(mmio_base, offset));
389 "RING_HEAD_ADDR = 0x%016"PRIx64,
390 fpga_reg_read_64(mmio_base, offset +
391 FPGA_LTE_FEC_RING_HEAD_ADDR));
393 "RING_SIZE = 0x%04"PRIx16,
394 fpga_reg_read_16(mmio_base, offset +
395 FPGA_LTE_FEC_RING_SIZE));
397 "RING_MISC = 0x%02"PRIx8,
398 fpga_reg_read_8(mmio_base, offset +
399 FPGA_LTE_FEC_RING_MISC));
401 "RING_ENABLE = 0x%02"PRIx8,
402 fpga_reg_read_8(mmio_base, offset +
403 FPGA_LTE_FEC_RING_ENABLE));
405 "RING_FLUSH_QUEUE_EN = 0x%02"PRIx8,
406 fpga_reg_read_8(mmio_base, offset +
407 FPGA_LTE_FEC_RING_FLUSH_QUEUE_EN));
409 "RING_SHADOW_TAIL = 0x%04"PRIx16,
410 fpga_reg_read_16(mmio_base, offset +
411 FPGA_LTE_FEC_RING_SHADOW_TAIL));
413 "RING_HEAD_POINT = 0x%04"PRIx16,
414 fpga_reg_read_16(mmio_base, offset +
415 FPGA_LTE_FEC_RING_HEAD_POINT));
418 /* Read Static Register of FPGA LTE FEC device */
420 print_static_reg_debug_info(void *mmio_base)
422 uint16_t config = fpga_reg_read_16(mmio_base,
423 FPGA_LTE_FEC_CONFIGURATION);
424 uint8_t qmap_done = fpga_reg_read_8(mmio_base,
425 FPGA_LTE_FEC_QUEUE_PF_VF_MAP_DONE);
426 uint16_t lb_factor = fpga_reg_read_16(mmio_base,
427 FPGA_LTE_FEC_LOAD_BALANCE_FACTOR);
428 uint16_t ring_desc_len = fpga_reg_read_16(mmio_base,
429 FPGA_LTE_FEC_RING_DESC_LEN);
430 uint16_t flr_time_out = fpga_reg_read_16(mmio_base,
431 FPGA_LTE_FEC_FLR_TIME_OUT);
433 rte_bbdev_log_debug("UL.DL Weights = %u.%u",
434 ((uint8_t)config), ((uint8_t)(config >> 8)));
435 rte_bbdev_log_debug("UL.DL Load Balance = %u.%u",
436 ((uint8_t)lb_factor), ((uint8_t)(lb_factor >> 8)));
437 rte_bbdev_log_debug("Queue-PF/VF Mapping Table = %s",
438 (qmap_done > 0) ? "READY" : "NOT-READY");
439 rte_bbdev_log_debug("Ring Descriptor Size = %u bytes",
440 ring_desc_len*FPGA_RING_DESC_LEN_UNIT_BYTES);
441 rte_bbdev_log_debug("FLR Timeout = %f usec",
442 (float)flr_time_out*FPGA_FLR_TIMEOUT_UNIT);
445 /* Print decode DMA Descriptor of FPGA LTE FEC device */
447 print_dma_dec_desc_debug_info(union fpga_dma_desc *desc)
449 rte_bbdev_log_debug("DMA response desc %p\n"
450 "\t-- done(%"PRIu32") | iter(%"PRIu32") | crc_pass(%"PRIu32")"
451 " | error (%"PRIu32") | crc_type(%"PRIu32")\n"
452 "\t-- max_iter(%"PRIu32") | bypass_rm(%"PRIu32") | "
453 "irq_en (%"PRIu32") | drop_crc(%"PRIu32") | offset(%"PRIu32")\n"
454 "\t-- k(%"PRIu32") | in_len (%"PRIu16") | op_add(%p)\n"
455 "\t-- cbs_in_op(%"PRIu32") | in_add (0x%08"PRIx32"%08"PRIx32") | "
456 "out_add (0x%08"PRIx32"%08"PRIx32")",
458 (uint32_t)desc->dec_req.done,
459 (uint32_t)desc->dec_req.iter,
460 (uint32_t)desc->dec_req.crc_pass,
461 (uint32_t)desc->dec_req.error,
462 (uint32_t)desc->dec_req.crc_type,
463 (uint32_t)desc->dec_req.max_iter,
464 (uint32_t)desc->dec_req.bypass_rm,
465 (uint32_t)desc->dec_req.irq_en,
466 (uint32_t)desc->dec_req.drop_crc,
467 (uint32_t)desc->dec_req.offset,
468 (uint32_t)desc->dec_req.k,
469 (uint16_t)desc->dec_req.in_len,
470 desc->dec_req.op_addr,
471 (uint32_t)desc->dec_req.cbs_in_op,
472 (uint32_t)desc->dec_req.in_addr_hi,
473 (uint32_t)desc->dec_req.in_addr_lw,
474 (uint32_t)desc->dec_req.out_addr_hi,
475 (uint32_t)desc->dec_req.out_addr_lw);
480 fpga_setup_queues(struct rte_bbdev *dev, uint16_t num_queues, int socket_id)
482 /* Number of queues bound to a PF/VF */
483 uint32_t hw_q_num = 0;
484 uint32_t ring_size, payload, address, q_id, offset;
485 rte_iova_t phys_addr;
486 struct fpga_ring_ctrl_reg ring_reg;
487 struct fpga_lte_fec_device *fpga_dev = dev->data->dev_private;
489 address = FPGA_LTE_FEC_QUEUE_PF_VF_MAP_DONE;
490 if (!(fpga_reg_read_32(fpga_dev->mmio_base, address) & 0x1)) {
492 "Queue-PF/VF mapping is not set! Was PF configured for device (%s) ?",
497 /* Clear queue registers structure */
498 memset(&ring_reg, 0, sizeof(struct fpga_ring_ctrl_reg));
501 * If a queue is valid and mapped to a calling PF/VF the read value is
502 * replaced with a queue ID and if it's not then
503 * FPGA_INVALID_HW_QUEUE_ID is returned.
505 for (q_id = 0; q_id < FPGA_TOTAL_NUM_QUEUES; ++q_id) {
506 uint32_t hw_q_id = fpga_reg_read_32(fpga_dev->mmio_base,
507 FPGA_LTE_FEC_QUEUE_MAP + (q_id << 2));
509 rte_bbdev_log_debug("%s: queue ID: %u, registry queue ID: %u",
510 dev->device->name, q_id, hw_q_id);
512 if (hw_q_id != FPGA_INVALID_HW_QUEUE_ID) {
513 fpga_dev->q_bound_bit_map |= (1ULL << q_id);
514 /* Clear queue register of found queue */
515 offset = FPGA_LTE_FEC_RING_CTRL_REGS +
516 (sizeof(struct fpga_ring_ctrl_reg) * q_id);
517 fpga_ring_reg_write(fpga_dev->mmio_base,
524 "No HW queues assigned to this device. Probably this is a VF configured for PF mode. Check device configuration!");
528 if (num_queues > hw_q_num) {
530 "Not enough queues for device %s! Requested: %u, available: %u",
531 dev->device->name, num_queues, hw_q_num);
535 ring_size = FPGA_RING_MAX_SIZE * sizeof(struct fpga_dma_dec_desc);
537 /* Enforce 32 byte alignment */
538 RTE_BUILD_BUG_ON((RTE_CACHE_LINE_SIZE % 32) != 0);
540 /* Allocate memory for SW descriptor rings */
541 fpga_dev->sw_rings = rte_zmalloc_socket(dev->device->driver->name,
542 num_queues * ring_size, RTE_CACHE_LINE_SIZE,
544 if (fpga_dev->sw_rings == NULL) {
546 "Failed to allocate memory for %s:%u sw_rings",
547 dev->device->driver->name, dev->data->dev_id);
551 fpga_dev->sw_rings_phys = rte_malloc_virt2iova(fpga_dev->sw_rings);
552 fpga_dev->sw_ring_size = ring_size;
553 fpga_dev->sw_ring_max_depth = FPGA_RING_MAX_SIZE;
555 /* Allocate memory for ring flush status */
556 fpga_dev->flush_queue_status = rte_zmalloc_socket(NULL,
557 sizeof(uint64_t), RTE_CACHE_LINE_SIZE, socket_id);
558 if (fpga_dev->flush_queue_status == NULL) {
560 "Failed to allocate memory for %s:%u flush_queue_status",
561 dev->device->driver->name, dev->data->dev_id);
565 /* Set the flush status address registers */
566 phys_addr = rte_malloc_virt2iova(fpga_dev->flush_queue_status);
568 address = FPGA_LTE_FEC_VFQ_FLUSH_STATUS_LW;
569 payload = (uint32_t)(phys_addr);
570 fpga_reg_write_32(fpga_dev->mmio_base, address, payload);
572 address = FPGA_LTE_FEC_VFQ_FLUSH_STATUS_HI;
573 payload = (uint32_t)(phys_addr >> 32);
574 fpga_reg_write_32(fpga_dev->mmio_base, address, payload);
580 fpga_dev_close(struct rte_bbdev *dev)
582 struct fpga_lte_fec_device *fpga_dev = dev->data->dev_private;
584 rte_free(fpga_dev->sw_rings);
585 rte_free(fpga_dev->flush_queue_status);
591 fpga_dev_info_get(struct rte_bbdev *dev,
592 struct rte_bbdev_driver_info *dev_info)
594 struct fpga_lte_fec_device *d = dev->data->dev_private;
597 /* TODO RTE_BBDEV_TURBO_NEG_LLR_1_BIT_IN and numbers of buffers are set
598 * to temporary values as they are required by test application while
601 static const struct rte_bbdev_op_cap bbdev_capabilities[] = {
603 .type = RTE_BBDEV_OP_TURBO_DEC,
606 RTE_BBDEV_TURBO_CRC_TYPE_24B |
607 RTE_BBDEV_TURBO_SUBBLOCK_DEINTERLEAVE |
608 RTE_BBDEV_TURBO_DEC_INTERRUPTS |
609 RTE_BBDEV_TURBO_NEG_LLR_1_BIT_IN |
610 RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP,
611 .max_llr_modulus = INT8_MAX,
613 RTE_BBDEV_TURBO_MAX_CODE_BLOCKS,
614 .num_buffers_hard_out =
615 RTE_BBDEV_TURBO_MAX_CODE_BLOCKS,
616 .num_buffers_soft_out = 0
620 .type = RTE_BBDEV_OP_TURBO_ENC,
623 RTE_BBDEV_TURBO_CRC_24B_ATTACH |
624 RTE_BBDEV_TURBO_RATE_MATCH |
625 RTE_BBDEV_TURBO_ENC_INTERRUPTS,
627 RTE_BBDEV_TURBO_MAX_CODE_BLOCKS,
629 RTE_BBDEV_TURBO_MAX_CODE_BLOCKS
632 RTE_BBDEV_END_OF_CAPABILITIES_LIST()
635 static struct rte_bbdev_queue_conf default_queue_conf;
636 default_queue_conf.socket = dev->data->socket_id;
637 default_queue_conf.queue_size = FPGA_RING_MAX_SIZE;
640 dev_info->driver_name = dev->device->driver->name;
641 dev_info->queue_size_lim = FPGA_RING_MAX_SIZE;
642 dev_info->hardware_accelerated = true;
643 dev_info->min_alignment = 64;
644 dev_info->default_queue_conf = default_queue_conf;
645 dev_info->capabilities = bbdev_capabilities;
646 dev_info->cpu_flag_reqs = NULL;
648 /* Calculates number of queues assigned to device */
649 dev_info->max_num_queues = 0;
650 for (q_id = 0; q_id < FPGA_TOTAL_NUM_QUEUES; ++q_id) {
651 uint32_t hw_q_id = fpga_reg_read_32(d->mmio_base,
652 FPGA_LTE_FEC_QUEUE_MAP + (q_id << 2));
653 if (hw_q_id != FPGA_INVALID_HW_QUEUE_ID)
654 dev_info->max_num_queues++;
659 * Find index of queue bound to current PF/VF which is unassigned. Return -1
660 * when there is no available queue
663 fpga_find_free_queue_idx(struct rte_bbdev *dev,
664 const struct rte_bbdev_queue_conf *conf)
666 struct fpga_lte_fec_device *d = dev->data->dev_private;
669 uint8_t range = FPGA_TOTAL_NUM_QUEUES >> 1;
671 if (conf->op_type == RTE_BBDEV_OP_TURBO_ENC) {
672 i = FPGA_NUM_DL_QUEUES;
673 range = FPGA_TOTAL_NUM_QUEUES;
676 for (; i < range; ++i) {
678 /* Check if index of queue is bound to current PF/VF */
679 if (d->q_bound_bit_map & q_idx)
680 /* Check if found queue was not already assigned */
681 if (!(d->q_assigned_bit_map & q_idx)) {
682 d->q_assigned_bit_map |= q_idx;
687 rte_bbdev_log(INFO, "Failed to find free queue on %s", dev->data->name);
693 fpga_queue_setup(struct rte_bbdev *dev, uint16_t queue_id,
694 const struct rte_bbdev_queue_conf *conf)
696 uint32_t address, ring_offset;
697 struct fpga_lte_fec_device *d = dev->data->dev_private;
698 struct fpga_queue *q;
701 /* Check if there is a free queue to assign */
702 q_idx = fpga_find_free_queue_idx(dev, conf);
706 /* Allocate the queue data structure. */
707 q = rte_zmalloc_socket(dev->device->driver->name, sizeof(*q),
708 RTE_CACHE_LINE_SIZE, conf->socket);
710 /* Mark queue as un-assigned */
711 d->q_assigned_bit_map &= (0xFFFFFFFF - (1ULL << q_idx));
712 rte_bbdev_log(ERR, "Failed to allocate queue memory");
719 /* Set ring_base_addr */
720 q->ring_addr = RTE_PTR_ADD(d->sw_rings, (d->sw_ring_size * queue_id));
721 q->ring_ctrl_reg.ring_base_addr = d->sw_rings_phys +
722 (d->sw_ring_size * queue_id);
724 /* Allocate memory for Completion Head variable*/
725 q->ring_head_addr = rte_zmalloc_socket(dev->device->driver->name,
726 sizeof(uint64_t), RTE_CACHE_LINE_SIZE, conf->socket);
727 if (q->ring_head_addr == NULL) {
728 /* Mark queue as un-assigned */
729 d->q_assigned_bit_map &= (0xFFFFFFFF - (1ULL << q_idx));
732 "Failed to allocate memory for %s:%u completion_head",
733 dev->device->driver->name, dev->data->dev_id);
736 /* Set ring_head_addr */
737 q->ring_ctrl_reg.ring_head_addr =
738 rte_malloc_virt2iova(q->ring_head_addr);
740 /* Clear shadow_completion_head */
741 q->shadow_completion_head = 0;
744 if (conf->queue_size > FPGA_RING_MAX_SIZE) {
745 /* Mark queue as un-assigned */
746 d->q_assigned_bit_map &= (0xFFFFFFFF - (1ULL << q_idx));
747 rte_free(q->ring_head_addr);
750 "Size of queue is too big %d (MAX: %d ) for %s:%u",
751 conf->queue_size, FPGA_RING_MAX_SIZE,
752 dev->device->driver->name, dev->data->dev_id);
755 q->ring_ctrl_reg.ring_size = conf->queue_size;
757 /* Set Miscellaneous FPGA register*/
758 /* Max iteration number for TTI mitigation - todo */
759 q->ring_ctrl_reg.max_ul_dec = 0;
760 /* Enable max iteration number for TTI - todo */
761 q->ring_ctrl_reg.max_ul_dec_en = 0;
763 /* Enable the ring */
764 q->ring_ctrl_reg.enable = 1;
766 /* Set FPGA head_point and tail registers */
767 q->ring_ctrl_reg.head_point = q->tail = 0;
769 /* Set FPGA shadow_tail register */
770 q->ring_ctrl_reg.shadow_tail = q->tail;
772 /* Calculates the ring offset for found queue */
773 ring_offset = FPGA_LTE_FEC_RING_CTRL_REGS +
774 (sizeof(struct fpga_ring_ctrl_reg) * q_idx);
776 /* Set FPGA Ring Control Registers */
777 fpga_ring_reg_write(d->mmio_base, ring_offset, q->ring_ctrl_reg);
779 /* Store MMIO register of shadow_tail */
780 address = ring_offset + FPGA_LTE_FEC_RING_SHADOW_TAIL;
781 q->shadow_tail_addr = RTE_PTR_ADD(d->mmio_base, address);
783 q->head_free_desc = q->tail;
786 q->sw_ring_wrap_mask = conf->queue_size - 1;
788 rte_bbdev_log_debug("Setup dev%u q%u: queue_idx=%u",
789 dev->data->dev_id, queue_id, q->q_idx);
791 dev->data->queues[queue_id].queue_private = q;
793 rte_bbdev_log_debug("BBDEV queue[%d] set up for FPGA queue[%d]",
796 #ifdef RTE_LIBRTE_BBDEV_DEBUG
797 /* Read FPGA Ring Control Registers after configuration*/
798 print_ring_reg_debug_info(d->mmio_base, ring_offset);
804 fpga_queue_release(struct rte_bbdev *dev, uint16_t queue_id)
806 struct fpga_lte_fec_device *d = dev->data->dev_private;
807 struct fpga_queue *q = dev->data->queues[queue_id].queue_private;
808 struct fpga_ring_ctrl_reg ring_reg;
811 rte_bbdev_log_debug("FPGA Queue[%d] released", queue_id);
814 memset(&ring_reg, 0, sizeof(struct fpga_ring_ctrl_reg));
815 offset = FPGA_LTE_FEC_RING_CTRL_REGS +
816 (sizeof(struct fpga_ring_ctrl_reg) * q->q_idx);
818 fpga_reg_write_8(d->mmio_base,
819 offset + FPGA_LTE_FEC_RING_ENABLE, 0x00);
820 /* Clear queue registers */
821 fpga_ring_reg_write(d->mmio_base, offset, ring_reg);
823 /* Mark the Queue as un-assigned */
824 d->q_assigned_bit_map &= (0xFFFFFFFF - (1ULL << q->q_idx));
825 rte_free(q->ring_head_addr);
827 dev->data->queues[queue_id].queue_private = NULL;
833 /* Function starts a device queue. */
835 fpga_queue_start(struct rte_bbdev *dev, uint16_t queue_id)
837 struct fpga_lte_fec_device *d = dev->data->dev_private;
838 #ifdef RTE_LIBRTE_BBDEV_DEBUG
840 rte_bbdev_log(ERR, "Invalid device pointer");
844 struct fpga_queue *q = dev->data->queues[queue_id].queue_private;
845 uint32_t offset = FPGA_LTE_FEC_RING_CTRL_REGS +
846 (sizeof(struct fpga_ring_ctrl_reg) * q->q_idx);
847 uint8_t enable = 0x01;
848 uint16_t zero = 0x0000;
850 /* Clear queue head and tail variables */
851 q->tail = q->head_free_desc = 0;
853 /* Clear FPGA head_point and tail registers */
854 fpga_reg_write_16(d->mmio_base, offset + FPGA_LTE_FEC_RING_HEAD_POINT,
856 fpga_reg_write_16(d->mmio_base, offset + FPGA_LTE_FEC_RING_SHADOW_TAIL,
860 fpga_reg_write_8(d->mmio_base, offset + FPGA_LTE_FEC_RING_ENABLE,
863 rte_bbdev_log_debug("FPGA Queue[%d] started", queue_id);
867 /* Function stops a device queue. */
869 fpga_queue_stop(struct rte_bbdev *dev, uint16_t queue_id)
871 struct fpga_lte_fec_device *d = dev->data->dev_private;
872 #ifdef RTE_LIBRTE_BBDEV_DEBUG
874 rte_bbdev_log(ERR, "Invalid device pointer");
878 struct fpga_queue *q = dev->data->queues[queue_id].queue_private;
879 uint32_t offset = FPGA_LTE_FEC_RING_CTRL_REGS +
880 (sizeof(struct fpga_ring_ctrl_reg) * q->q_idx);
881 uint8_t payload = 0x01;
883 uint8_t timeout = FPGA_QUEUE_FLUSH_TIMEOUT_US /
884 FPGA_TIMEOUT_CHECK_INTERVAL;
886 /* Set flush_queue_en bit to trigger queue flushing */
887 fpga_reg_write_8(d->mmio_base,
888 offset + FPGA_LTE_FEC_RING_FLUSH_QUEUE_EN, payload);
890 /** Check if queue flush is completed.
891 * FPGA will update the completion flag after queue flushing is
892 * completed. If completion flag is not updated within 1ms it is
893 * considered as a failure.
895 while (!(*((volatile uint8_t *)d->flush_queue_status + q->q_idx) & payload)) {
896 if (counter > timeout) {
897 rte_bbdev_log(ERR, "FPGA Queue Flush failed for queue %d",
901 usleep(FPGA_TIMEOUT_CHECK_INTERVAL);
907 fpga_reg_write_8(d->mmio_base, offset + FPGA_LTE_FEC_RING_ENABLE,
910 rte_bbdev_log_debug("FPGA Queue[%d] stopped", queue_id);
914 static inline uint16_t
915 get_queue_id(struct rte_bbdev_data *data, uint8_t q_idx)
919 for (queue_id = 0; queue_id < data->num_queues; ++queue_id) {
920 struct fpga_queue *q = data->queues[queue_id].queue_private;
921 if (q != NULL && q->q_idx == q_idx)
928 /* Interrupt handler triggered by FPGA dev for handling specific interrupt */
930 fpga_dev_interrupt_handler(void *cb_arg)
932 struct rte_bbdev *dev = cb_arg;
933 struct fpga_lte_fec_device *fpga_dev = dev->data->dev_private;
934 struct fpga_queue *q;
940 /* Scan queue assigned to this device */
941 for (i = 0; i < FPGA_TOTAL_NUM_QUEUES; ++i) {
943 if (fpga_dev->q_bound_bit_map & q_idx) {
944 queue_id = get_queue_id(dev->data, i);
945 if (queue_id == (uint16_t) -1)
948 /* Check if completion head was changed */
949 q = dev->data->queues[queue_id].queue_private;
950 ring_head = *q->ring_head_addr;
951 if (q->shadow_completion_head != ring_head &&
952 q->irq_enable == 1) {
953 q->shadow_completion_head = ring_head;
954 rte_bbdev_pmd_callback_process(
956 RTE_BBDEV_EVENT_DEQUEUE,
964 fpga_queue_intr_enable(struct rte_bbdev *dev, uint16_t queue_id)
966 struct fpga_queue *q = dev->data->queues[queue_id].queue_private;
968 if (!rte_intr_cap_multiple(dev->intr_handle))
977 fpga_queue_intr_disable(struct rte_bbdev *dev, uint16_t queue_id)
979 struct fpga_queue *q = dev->data->queues[queue_id].queue_private;
986 fpga_intr_enable(struct rte_bbdev *dev)
991 if (!rte_intr_cap_multiple(dev->intr_handle)) {
992 rte_bbdev_log(ERR, "Multiple intr vector is not supported by FPGA (%s)",
997 /* Create event file descriptors for each of 64 queue. Event fds will be
998 * mapped to FPGA IRQs in rte_intr_enable(). This is a 1:1 mapping where
999 * the IRQ number is a direct translation to the queue number.
1001 * 63 (FPGA_NUM_INTR_VEC) event fds are created as rte_intr_enable()
1002 * mapped the first IRQ to already created interrupt event file
1003 * descriptor (intr_handle->fd).
1005 if (rte_intr_efd_enable(dev->intr_handle, FPGA_NUM_INTR_VEC)) {
1006 rte_bbdev_log(ERR, "Failed to create fds for %u queues",
1007 dev->data->num_queues);
1011 /* TODO Each event file descriptor is overwritten by interrupt event
1012 * file descriptor. That descriptor is added to epoll observed list.
1013 * It ensures that callback function assigned to that descriptor will
1014 * invoked when any FPGA queue issues interrupt.
1016 for (i = 0; i < FPGA_NUM_INTR_VEC; ++i)
1017 dev->intr_handle->efds[i] = dev->intr_handle->fd;
1019 if (!dev->intr_handle->intr_vec) {
1020 dev->intr_handle->intr_vec = rte_zmalloc("intr_vec",
1021 dev->data->num_queues * sizeof(int), 0);
1022 if (!dev->intr_handle->intr_vec) {
1023 rte_bbdev_log(ERR, "Failed to allocate %u vectors",
1024 dev->data->num_queues);
1029 ret = rte_intr_enable(dev->intr_handle);
1032 "Couldn't enable interrupts for device: %s",
1037 ret = rte_intr_callback_register(dev->intr_handle,
1038 fpga_dev_interrupt_handler, dev);
1041 "Couldn't register interrupt callback for device: %s",
1049 static const struct rte_bbdev_ops fpga_ops = {
1050 .setup_queues = fpga_setup_queues,
1051 .intr_enable = fpga_intr_enable,
1052 .close = fpga_dev_close,
1053 .info_get = fpga_dev_info_get,
1054 .queue_setup = fpga_queue_setup,
1055 .queue_stop = fpga_queue_stop,
1056 .queue_start = fpga_queue_start,
1057 .queue_release = fpga_queue_release,
1058 .queue_intr_enable = fpga_queue_intr_enable,
1059 .queue_intr_disable = fpga_queue_intr_disable
1063 fpga_dma_enqueue(struct fpga_queue *q, uint16_t num_desc,
1064 struct rte_bbdev_stats *queue_stats)
1066 #ifdef RTE_BBDEV_OFFLOAD_COST
1067 uint64_t start_time = 0;
1068 queue_stats->acc_offload_cycles = 0;
1070 RTE_SET_USED(queue_stats);
1073 /* Update tail and shadow_tail register */
1074 q->tail = (q->tail + num_desc) & q->sw_ring_wrap_mask;
1078 #ifdef RTE_BBDEV_OFFLOAD_COST
1079 /* Start time measurement for enqueue function offload. */
1080 start_time = rte_rdtsc_precise();
1082 mmio_write_16(q->shadow_tail_addr, q->tail);
1084 #ifdef RTE_BBDEV_OFFLOAD_COST
1086 queue_stats->acc_offload_cycles += rte_rdtsc_precise() - start_time;
1090 /* Calculates number of CBs in processed encoder TB based on 'r' and input
1093 static inline uint8_t
1094 get_num_cbs_in_op_enc(struct rte_bbdev_op_turbo_enc *turbo_enc)
1096 uint8_t c, c_neg, r, crc24_bits = 0;
1097 uint16_t k, k_neg, k_pos;
1098 uint8_t cbs_in_op = 0;
1101 length = turbo_enc->input.length;
1102 r = turbo_enc->tb_params.r;
1103 c = turbo_enc->tb_params.c;
1104 c_neg = turbo_enc->tb_params.c_neg;
1105 k_neg = turbo_enc->tb_params.k_neg;
1106 k_pos = turbo_enc->tb_params.k_pos;
1108 while (length > 0 && r < c) {
1109 k = (r < c_neg) ? k_neg : k_pos;
1110 length -= (k - crc24_bits) >> 3;
1118 /* Calculates number of CBs in processed decoder TB based on 'r' and input
1121 static inline uint16_t
1122 get_num_cbs_in_op_dec(struct rte_bbdev_op_turbo_dec *turbo_dec)
1124 uint8_t c, c_neg, r = 0;
1125 uint16_t kw, k, k_neg, k_pos, cbs_in_op = 0;
1128 length = turbo_dec->input.length;
1129 r = turbo_dec->tb_params.r;
1130 c = turbo_dec->tb_params.c;
1131 c_neg = turbo_dec->tb_params.c_neg;
1132 k_neg = turbo_dec->tb_params.k_neg;
1133 k_pos = turbo_dec->tb_params.k_pos;
1134 while (length > 0 && r < c) {
1135 k = (r < c_neg) ? k_neg : k_pos;
1136 kw = RTE_ALIGN_CEIL(k + 4, 32) * 3;
1145 /* Read flag value 0/1/ from bitmap */
1147 check_bit(uint32_t bitmap, uint32_t bitmask)
1149 return bitmap & bitmask;
1152 /* Print an error if a descriptor error has occurred.
1153 * Return 0 on success, 1 on failure
1156 check_desc_error(uint32_t error_code) {
1157 switch (error_code) {
1158 case DESC_ERR_NO_ERR:
1160 case DESC_ERR_K_OUT_OF_RANGE:
1161 rte_bbdev_log(ERR, "Block_size_k is out of range (k<40 or k>6144)");
1163 case DESC_ERR_K_NOT_NORMAL:
1164 rte_bbdev_log(ERR, "Block_size_k is not a normal value within normal range");
1166 case DESC_ERR_KPAI_NOT_NORMAL:
1167 rte_bbdev_log(ERR, "Three_kpai is not a normal value for UL only");
1169 case DESC_ERR_DESC_OFFSET_ERR:
1170 rte_bbdev_log(ERR, "Queue offset does not meet the expectation in the FPGA");
1172 case (DESC_ERR_K_OUT_OF_RANGE | DESC_ERR_DESC_OFFSET_ERR):
1173 rte_bbdev_log(ERR, "Block_size_k is out of range (k<40 or k>6144) and queue offset error");
1175 case (DESC_ERR_K_NOT_NORMAL | DESC_ERR_DESC_OFFSET_ERR):
1176 rte_bbdev_log(ERR, "Block_size_k is not a normal value within normal range and queue offset error");
1178 case (DESC_ERR_KPAI_NOT_NORMAL | DESC_ERR_DESC_OFFSET_ERR):
1179 rte_bbdev_log(ERR, "Three_kpai is not a normal value for UL only and queue offset error");
1181 case DESC_ERR_DESC_READ_FAIL:
1182 rte_bbdev_log(ERR, "Unsuccessful completion for descriptor read");
1184 case DESC_ERR_DESC_READ_TIMEOUT:
1185 rte_bbdev_log(ERR, "Descriptor read time-out");
1187 case DESC_ERR_DESC_READ_TLP_POISONED:
1188 rte_bbdev_log(ERR, "Descriptor read TLP poisoned");
1190 case DESC_ERR_CB_READ_FAIL:
1191 rte_bbdev_log(ERR, "Unsuccessful completion for code block");
1193 case DESC_ERR_CB_READ_TIMEOUT:
1194 rte_bbdev_log(ERR, "Code block read time-out");
1196 case DESC_ERR_CB_READ_TLP_POISONED:
1197 rte_bbdev_log(ERR, "Code block read TLP poisoned");
1200 rte_bbdev_log(ERR, "Descriptor error unknown error code %u",
1208 * Set DMA descriptor for encode operation (1 Code Block)
1211 * Pointer to a single encode operation.
1213 * Pointer to DMA descriptor.
1215 * Pointer to pointer to input data which will be decoded.
1217 * K value (length of input in bits).
1219 * E value (length of output in bits).
1221 * Ncb value (size of the soft buffer).
1223 * Length of output buffer
1225 * Input offset in rte_mbuf structure. It is used for calculating the point
1226 * where data is starting.
1228 * Output offset in rte_mbuf structure. It is used for calculating the point
1229 * where hard output data will be stored.
1231 * Number of CBs contained in one operation.
1234 fpga_dma_desc_te_fill(struct rte_bbdev_enc_op *op,
1235 struct fpga_dma_enc_desc *desc, struct rte_mbuf *input,
1236 struct rte_mbuf *output, uint16_t k, uint16_t e, uint16_t ncb,
1237 uint32_t in_offset, uint32_t out_offset, uint16_t desc_offset,
1243 desc->crc_en = check_bit(op->turbo_enc.op_flags,
1244 RTE_BBDEV_TURBO_CRC_24B_ATTACH);
1245 desc->bypass_rm = !check_bit(op->turbo_enc.op_flags,
1246 RTE_BBDEV_TURBO_RATE_MATCH);
1250 desc->rv = op->turbo_enc.rv_index;
1251 desc->offset = desc_offset;
1252 /* Set inbound data buffer address */
1253 desc->in_addr_hi = (uint32_t)(
1254 rte_pktmbuf_iova_offset(input, in_offset) >> 32);
1255 desc->in_addr_lw = (uint32_t)(
1256 rte_pktmbuf_iova_offset(input, in_offset));
1258 desc->out_addr_hi = (uint32_t)(
1259 rte_pktmbuf_iova_offset(output, out_offset) >> 32);
1260 desc->out_addr_lw = (uint32_t)(
1261 rte_pktmbuf_iova_offset(output, out_offset));
1263 /* Save software context needed for dequeue */
1266 /* Set total number of CBs in an op */
1267 desc->cbs_in_op = cbs_in_op;
1273 * Set DMA descriptor for encode operation (1 Code Block)
1276 * Pointer to a single encode operation.
1278 * Pointer to DMA descriptor.
1280 * Pointer to pointer to input data which will be decoded.
1282 * Length of an input.
1284 * K value (length of an output in bits).
1286 * Input offset in rte_mbuf structure. It is used for calculating the point
1287 * where data is starting.
1289 * Output offset in rte_mbuf structure. It is used for calculating the point
1290 * where hard output data will be stored.
1292 * Number of CBs contained in one operation.
1295 fpga_dma_desc_td_fill(struct rte_bbdev_dec_op *op,
1296 struct fpga_dma_dec_desc *desc, struct rte_mbuf *input,
1297 struct rte_mbuf *output, uint16_t in_length, uint16_t k,
1298 uint32_t in_offset, uint32_t out_offset, uint16_t desc_offset,
1303 /* Set inbound data buffer address */
1304 desc->in_addr_hi = (uint32_t)(
1305 rte_pktmbuf_iova_offset(input, in_offset) >> 32);
1306 desc->in_addr_lw = (uint32_t)(
1307 rte_pktmbuf_iova_offset(input, in_offset));
1308 desc->in_len = in_length;
1310 desc->crc_type = !check_bit(op->turbo_dec.op_flags,
1311 RTE_BBDEV_TURBO_CRC_TYPE_24B);
1312 if ((op->turbo_dec.code_block_mode == 0)
1313 && !check_bit(op->turbo_dec.op_flags,
1314 RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP))
1316 desc->max_iter = op->turbo_dec.iter_max * 2;
1317 desc->offset = desc_offset;
1318 desc->out_addr_hi = (uint32_t)(
1319 rte_pktmbuf_iova_offset(output, out_offset) >> 32);
1320 desc->out_addr_lw = (uint32_t)(
1321 rte_pktmbuf_iova_offset(output, out_offset));
1323 /* Save software context needed for dequeue */
1326 /* Set total number of CBs in an op */
1327 desc->cbs_in_op = cbs_in_op;
1332 #ifdef RTE_LIBRTE_BBDEV_DEBUG
1333 /* Validates turbo encoder parameters */
1335 validate_enc_op(struct rte_bbdev_enc_op *op)
1337 struct rte_bbdev_op_turbo_enc *turbo_enc = &op->turbo_enc;
1338 struct rte_bbdev_op_enc_turbo_cb_params *cb = NULL;
1339 struct rte_bbdev_op_enc_turbo_tb_params *tb = NULL;
1340 uint16_t kw, kw_neg, kw_pos;
1342 if (turbo_enc->input.length >
1343 RTE_BBDEV_TURBO_MAX_TB_SIZE >> 3) {
1344 rte_bbdev_log(ERR, "TB size (%u) is too big, max: %d",
1345 turbo_enc->input.length,
1346 RTE_BBDEV_TURBO_MAX_TB_SIZE);
1347 op->status = 1 << RTE_BBDEV_DATA_ERROR;
1351 if (op->mempool == NULL) {
1352 rte_bbdev_log(ERR, "Invalid mempool pointer");
1355 if (turbo_enc->input.data == NULL) {
1356 rte_bbdev_log(ERR, "Invalid input pointer");
1359 if (turbo_enc->output.data == NULL) {
1360 rte_bbdev_log(ERR, "Invalid output pointer");
1363 if (turbo_enc->rv_index > 3) {
1365 "rv_index (%u) is out of range 0 <= value <= 3",
1366 turbo_enc->rv_index);
1369 if (turbo_enc->code_block_mode != 0 &&
1370 turbo_enc->code_block_mode != 1) {
1372 "code_block_mode (%u) is out of range 0 <= value <= 1",
1373 turbo_enc->code_block_mode);
1377 if (turbo_enc->code_block_mode == 0) {
1378 tb = &turbo_enc->tb_params;
1379 if ((tb->k_neg < RTE_BBDEV_TURBO_MIN_CB_SIZE
1380 || tb->k_neg > RTE_BBDEV_TURBO_MAX_CB_SIZE)
1383 "k_neg (%u) is out of range %u <= value <= %u",
1384 tb->k_neg, RTE_BBDEV_TURBO_MIN_CB_SIZE,
1385 RTE_BBDEV_TURBO_MAX_CB_SIZE);
1388 if (tb->k_pos < RTE_BBDEV_TURBO_MIN_CB_SIZE
1389 || tb->k_pos > RTE_BBDEV_TURBO_MAX_CB_SIZE) {
1391 "k_pos (%u) is out of range %u <= value <= %u",
1392 tb->k_pos, RTE_BBDEV_TURBO_MIN_CB_SIZE,
1393 RTE_BBDEV_TURBO_MAX_CB_SIZE);
1396 if (tb->c_neg > (RTE_BBDEV_TURBO_MAX_CODE_BLOCKS - 1))
1398 "c_neg (%u) is out of range 0 <= value <= %u",
1400 RTE_BBDEV_TURBO_MAX_CODE_BLOCKS - 1);
1401 if (tb->c < 1 || tb->c > RTE_BBDEV_TURBO_MAX_CODE_BLOCKS) {
1403 "c (%u) is out of range 1 <= value <= %u",
1404 tb->c, RTE_BBDEV_TURBO_MAX_CODE_BLOCKS);
1407 if (tb->cab > tb->c) {
1409 "cab (%u) is greater than c (%u)",
1413 if ((tb->ea < RTE_BBDEV_TURBO_MIN_CB_SIZE || (tb->ea % 2))
1414 && tb->r < tb->cab) {
1416 "ea (%u) is less than %u or it is not even",
1417 tb->ea, RTE_BBDEV_TURBO_MIN_CB_SIZE);
1420 if ((tb->eb < RTE_BBDEV_TURBO_MIN_CB_SIZE || (tb->eb % 2))
1421 && tb->c > tb->cab) {
1423 "eb (%u) is less than %u or it is not even",
1424 tb->eb, RTE_BBDEV_TURBO_MIN_CB_SIZE);
1428 kw_neg = 3 * RTE_ALIGN_CEIL(tb->k_neg + 4,
1429 RTE_BBDEV_TURBO_C_SUBBLOCK);
1430 if (tb->ncb_neg < tb->k_neg || tb->ncb_neg > kw_neg) {
1432 "ncb_neg (%u) is out of range (%u) k_neg <= value <= (%u) kw_neg",
1433 tb->ncb_neg, tb->k_neg, kw_neg);
1437 kw_pos = 3 * RTE_ALIGN_CEIL(tb->k_pos + 4,
1438 RTE_BBDEV_TURBO_C_SUBBLOCK);
1439 if (tb->ncb_pos < tb->k_pos || tb->ncb_pos > kw_pos) {
1441 "ncb_pos (%u) is out of range (%u) k_pos <= value <= (%u) kw_pos",
1442 tb->ncb_pos, tb->k_pos, kw_pos);
1445 if (tb->r > (tb->c - 1)) {
1447 "r (%u) is greater than c - 1 (%u)",
1452 cb = &turbo_enc->cb_params;
1453 if (cb->k < RTE_BBDEV_TURBO_MIN_CB_SIZE
1454 || cb->k > RTE_BBDEV_TURBO_MAX_CB_SIZE) {
1456 "k (%u) is out of range %u <= value <= %u",
1457 cb->k, RTE_BBDEV_TURBO_MIN_CB_SIZE,
1458 RTE_BBDEV_TURBO_MAX_CB_SIZE);
1462 if (cb->e < RTE_BBDEV_TURBO_MIN_CB_SIZE || (cb->e % 2)) {
1464 "e (%u) is less than %u or it is not even",
1465 cb->e, RTE_BBDEV_TURBO_MIN_CB_SIZE);
1469 kw = RTE_ALIGN_CEIL(cb->k + 4, RTE_BBDEV_TURBO_C_SUBBLOCK) * 3;
1470 if (cb->ncb < cb->k || cb->ncb > kw) {
1472 "ncb (%u) is out of range (%u) k <= value <= (%u) kw",
1473 cb->ncb, cb->k, kw);
1482 static inline char *
1483 mbuf_append(struct rte_mbuf *m_head, struct rte_mbuf *m, uint16_t len)
1485 if (unlikely(len > rte_pktmbuf_tailroom(m)))
1488 char *tail = (char *)m->buf_addr + m->data_off + m->data_len;
1489 m->data_len = (uint16_t)(m->data_len + len);
1490 m_head->pkt_len = (m_head->pkt_len + len);
1495 enqueue_enc_one_op_cb(struct fpga_queue *q, struct rte_bbdev_enc_op *op,
1496 uint16_t desc_offset)
1498 union fpga_dma_desc *desc;
1499 struct rte_mbuf *input;
1500 struct rte_mbuf *output;
1502 uint16_t k, e, ncb, ring_offset;
1503 uint32_t total_left, in_length, out_length, in_offset, out_offset;
1505 #ifdef RTE_LIBRTE_BBDEV_DEBUG
1506 /* Validate op structure */
1507 if (validate_enc_op(op) == -1) {
1508 rte_bbdev_log(ERR, "Turbo encoder validation failed");
1513 input = op->turbo_enc.input.data;
1514 output = op->turbo_enc.output.data;
1515 in_offset = op->turbo_enc.input.offset;
1516 out_offset = op->turbo_enc.output.offset;
1517 total_left = op->turbo_enc.input.length;
1518 k = op->turbo_enc.cb_params.k;
1519 e = op->turbo_enc.cb_params.e;
1520 ncb = op->turbo_enc.cb_params.ncb;
1522 if (check_bit(op->turbo_enc.op_flags, RTE_BBDEV_TURBO_CRC_24B_ATTACH))
1523 in_length = ((k - 24) >> 3);
1527 if (check_bit(op->turbo_enc.op_flags, RTE_BBDEV_TURBO_RATE_MATCH))
1528 out_length = (e + 7) >> 3;
1530 out_length = (k >> 3) * 3 + 2;
1532 mbuf_append(output, output, out_length);
1534 /* Offset into the ring */
1535 ring_offset = ((q->tail + desc_offset) & q->sw_ring_wrap_mask);
1536 /* Setup DMA Descriptor */
1537 desc = q->ring_addr + ring_offset;
1539 ret = fpga_dma_desc_te_fill(op, &desc->enc_req, input, output, k, e,
1540 ncb, in_offset, out_offset, ring_offset, 1);
1541 if (unlikely(ret < 0))
1544 /* Update lengths */
1545 total_left -= in_length;
1546 op->turbo_enc.output.length += out_length;
1548 if (total_left > 0) {
1550 "Mismatch between mbuf length and included CB sizes: mbuf len %u, cb len %u",
1551 total_left, in_length);
1559 enqueue_enc_one_op_tb(struct fpga_queue *q, struct rte_bbdev_enc_op *op,
1560 uint16_t desc_offset, uint8_t cbs_in_op)
1562 union fpga_dma_desc *desc;
1563 struct rte_mbuf *input, *output_head, *output;
1565 uint8_t r, c, crc24_bits = 0;
1566 uint16_t k, e, ncb, ring_offset;
1567 uint32_t mbuf_total_left, in_length, out_length, in_offset, out_offset;
1568 uint32_t seg_total_left;
1569 uint16_t current_enqueued_cbs = 0;
1571 #ifdef RTE_LIBRTE_BBDEV_DEBUG
1572 /* Validate op structure */
1573 if (validate_enc_op(op) == -1) {
1574 rte_bbdev_log(ERR, "Turbo encoder validation failed");
1579 input = op->turbo_enc.input.data;
1580 output_head = output = op->turbo_enc.output.data;
1581 in_offset = op->turbo_enc.input.offset;
1582 out_offset = op->turbo_enc.output.offset;
1583 mbuf_total_left = op->turbo_enc.input.length;
1585 c = op->turbo_enc.tb_params.c;
1586 r = op->turbo_enc.tb_params.r;
1588 if (check_bit(op->turbo_enc.op_flags, RTE_BBDEV_TURBO_CRC_24B_ATTACH))
1591 while (mbuf_total_left > 0 && r < c && input != NULL) {
1592 seg_total_left = rte_pktmbuf_data_len(input) - in_offset;
1594 e = (r < op->turbo_enc.tb_params.cab) ?
1595 op->turbo_enc.tb_params.ea :
1596 op->turbo_enc.tb_params.eb;
1597 k = (r < op->turbo_enc.tb_params.c_neg) ?
1598 op->turbo_enc.tb_params.k_neg :
1599 op->turbo_enc.tb_params.k_pos;
1600 ncb = (r < op->turbo_enc.tb_params.c_neg) ?
1601 op->turbo_enc.tb_params.ncb_neg :
1602 op->turbo_enc.tb_params.ncb_pos;
1604 in_length = ((k - crc24_bits) >> 3);
1606 if (check_bit(op->turbo_enc.op_flags,
1607 RTE_BBDEV_TURBO_RATE_MATCH))
1608 out_length = (e + 7) >> 3;
1610 out_length = (k >> 3) * 3 + 2;
1612 mbuf_append(output_head, output, out_length);
1614 /* Setup DMA Descriptor */
1615 ring_offset = ((q->tail + desc_offset) & q->sw_ring_wrap_mask);
1616 desc = q->ring_addr + ring_offset;
1617 ret = fpga_dma_desc_te_fill(op, &desc->enc_req, input, output,
1618 k, e, ncb, in_offset, out_offset, ring_offset,
1620 if (unlikely(ret < 0))
1623 rte_bbdev_log_debug("DMA request desc %p", desc);
1625 /* Update lengths */
1626 op->turbo_enc.output.length += out_length;
1627 mbuf_total_left -= in_length;
1629 /* Update offsets */
1630 if (seg_total_left == in_length) {
1631 /* Go to the next mbuf */
1632 input = input->next;
1633 output = output->next;
1637 in_offset += in_length;
1638 out_offset += out_length;
1643 current_enqueued_cbs++;
1646 if (mbuf_total_left > 0) {
1648 "Some date still left for processing: mbuf_total_left = %u",
1653 return current_enqueued_cbs;
1656 #ifdef RTE_LIBRTE_BBDEV_DEBUG
1657 /* Validates turbo decoder parameters */
1659 validate_dec_op(struct rte_bbdev_dec_op *op)
1661 struct rte_bbdev_op_turbo_dec *turbo_dec = &op->turbo_dec;
1662 struct rte_bbdev_op_dec_turbo_cb_params *cb = NULL;
1663 struct rte_bbdev_op_dec_turbo_tb_params *tb = NULL;
1665 if (op->mempool == NULL) {
1666 rte_bbdev_log(ERR, "Invalid mempool pointer");
1669 if (turbo_dec->input.data == NULL) {
1670 rte_bbdev_log(ERR, "Invalid input pointer");
1673 if (turbo_dec->hard_output.data == NULL) {
1674 rte_bbdev_log(ERR, "Invalid hard_output pointer");
1677 if (turbo_dec->rv_index > 3) {
1679 "rv_index (%u) is out of range 0 <= value <= 3",
1680 turbo_dec->rv_index);
1683 if (turbo_dec->iter_min < 1) {
1685 "iter_min (%u) is less than 1",
1686 turbo_dec->iter_min);
1689 if (turbo_dec->iter_max <= 2) {
1691 "iter_max (%u) is less than or equal to 2",
1692 turbo_dec->iter_max);
1695 if (turbo_dec->iter_min > turbo_dec->iter_max) {
1697 "iter_min (%u) is greater than iter_max (%u)",
1698 turbo_dec->iter_min, turbo_dec->iter_max);
1701 if (turbo_dec->code_block_mode != 0 &&
1702 turbo_dec->code_block_mode != 1) {
1704 "code_block_mode (%u) is out of range 0 <= value <= 1",
1705 turbo_dec->code_block_mode);
1709 if (turbo_dec->code_block_mode == 0) {
1711 if ((turbo_dec->op_flags &
1712 RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP) &&
1713 !(turbo_dec->op_flags & RTE_BBDEV_TURBO_CRC_TYPE_24B)) {
1715 "RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP should accompany RTE_BBDEV_TURBO_CRC_TYPE_24B");
1719 tb = &turbo_dec->tb_params;
1720 if ((tb->k_neg < RTE_BBDEV_TURBO_MIN_CB_SIZE
1721 || tb->k_neg > RTE_BBDEV_TURBO_MAX_CB_SIZE)
1724 "k_neg (%u) is out of range %u <= value <= %u",
1725 tb->k_neg, RTE_BBDEV_TURBO_MIN_CB_SIZE,
1726 RTE_BBDEV_TURBO_MAX_CB_SIZE);
1729 if ((tb->k_pos < RTE_BBDEV_TURBO_MIN_CB_SIZE
1730 || tb->k_pos > RTE_BBDEV_TURBO_MAX_CB_SIZE)
1731 && tb->c > tb->c_neg) {
1733 "k_pos (%u) is out of range %u <= value <= %u",
1734 tb->k_pos, RTE_BBDEV_TURBO_MIN_CB_SIZE,
1735 RTE_BBDEV_TURBO_MAX_CB_SIZE);
1738 if (tb->c_neg > (RTE_BBDEV_TURBO_MAX_CODE_BLOCKS - 1))
1740 "c_neg (%u) is out of range 0 <= value <= %u",
1742 RTE_BBDEV_TURBO_MAX_CODE_BLOCKS - 1);
1743 if (tb->c < 1 || tb->c > RTE_BBDEV_TURBO_MAX_CODE_BLOCKS) {
1745 "c (%u) is out of range 1 <= value <= %u",
1746 tb->c, RTE_BBDEV_TURBO_MAX_CODE_BLOCKS);
1749 if (tb->cab > tb->c) {
1751 "cab (%u) is greater than c (%u)",
1757 if (turbo_dec->op_flags & RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP) {
1759 "RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP is invalid in CB-mode");
1763 cb = &turbo_dec->cb_params;
1764 if (cb->k < RTE_BBDEV_TURBO_MIN_CB_SIZE
1765 || cb->k > RTE_BBDEV_TURBO_MAX_CB_SIZE) {
1767 "k (%u) is out of range %u <= value <= %u",
1768 cb->k, RTE_BBDEV_TURBO_MIN_CB_SIZE,
1769 RTE_BBDEV_TURBO_MAX_CB_SIZE);
1779 enqueue_dec_one_op_cb(struct fpga_queue *q, struct rte_bbdev_dec_op *op,
1780 uint16_t desc_offset)
1782 union fpga_dma_desc *desc;
1783 struct rte_mbuf *input;
1784 struct rte_mbuf *output;
1786 uint16_t k, kw, ring_offset;
1787 uint32_t total_left, in_length, out_length, in_offset, out_offset;
1789 #ifdef RTE_LIBRTE_BBDEV_DEBUG
1790 /* Validate op structure */
1791 if (validate_dec_op(op) == -1) {
1792 rte_bbdev_log(ERR, "Turbo decoder validation failed");
1797 input = op->turbo_dec.input.data;
1798 output = op->turbo_dec.hard_output.data;
1799 total_left = op->turbo_dec.input.length;
1800 in_offset = op->turbo_dec.input.offset;
1801 out_offset = op->turbo_dec.hard_output.offset;
1803 k = op->turbo_dec.cb_params.k;
1804 kw = RTE_ALIGN_CEIL(k + 4, 32) * 3;
1806 out_length = k >> 3;
1808 mbuf_append(output, output, out_length);
1810 /* Setup DMA Descriptor */
1811 ring_offset = ((q->tail + desc_offset) & q->sw_ring_wrap_mask);
1812 desc = q->ring_addr + ring_offset;
1813 ret = fpga_dma_desc_td_fill(op, &desc->dec_req, input, output,
1814 in_length, k, in_offset, out_offset, ring_offset, 1);
1815 if (unlikely(ret < 0))
1818 #ifdef RTE_LIBRTE_BBDEV_DEBUG
1819 print_dma_dec_desc_debug_info(desc);
1822 /* Update lengths */
1823 total_left -= in_length;
1824 op->turbo_dec.hard_output.length += out_length;
1826 if (total_left > 0) {
1828 "Mismatch between mbuf length and included CB sizes: mbuf len %u, cb len %u",
1829 total_left, in_length);
1838 enqueue_dec_one_op_tb(struct fpga_queue *q, struct rte_bbdev_dec_op *op,
1839 uint16_t desc_offset, uint8_t cbs_in_op)
1841 union fpga_dma_desc *desc;
1842 struct rte_mbuf *input, *output_head, *output;
1845 uint16_t k, kw, in_length, out_length, ring_offset;
1846 uint32_t mbuf_total_left, seg_total_left, in_offset, out_offset;
1847 uint16_t current_enqueued_cbs = 0;
1848 uint16_t crc24_overlap = 0;
1850 #ifdef RTE_LIBRTE_BBDEV_DEBUG
1851 /* Validate op structure */
1852 if (validate_dec_op(op) == -1) {
1853 rte_bbdev_log(ERR, "Turbo decoder validation failed");
1858 input = op->turbo_dec.input.data;
1859 output_head = output = op->turbo_dec.hard_output.data;
1860 mbuf_total_left = op->turbo_dec.input.length;
1861 in_offset = op->turbo_dec.input.offset;
1862 out_offset = op->turbo_dec.hard_output.offset;
1864 if (!check_bit(op->turbo_dec.op_flags,
1865 RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP))
1868 c = op->turbo_dec.tb_params.c;
1869 r = op->turbo_dec.tb_params.r;
1871 while (mbuf_total_left > 0 && r < c && input != NULL) {
1872 seg_total_left = rte_pktmbuf_data_len(input) - in_offset;
1873 k = (r < op->turbo_dec.tb_params.c_neg) ?
1874 op->turbo_dec.tb_params.k_neg :
1875 op->turbo_dec.tb_params.k_pos;
1876 kw = RTE_ALIGN_CEIL(k + 4, 32) * 3;
1879 out_length = (k - crc24_overlap) >> 3;
1881 mbuf_append(output_head, output, out_length);
1883 if (seg_total_left < in_length) {
1885 "Partial CB found in a TB. FPGA Driver doesn't support scatter-gather operations!");
1889 /* Setup DMA Descriptor */
1890 ring_offset = ((q->tail + desc_offset) & q->sw_ring_wrap_mask);
1891 desc = q->ring_addr + ring_offset;
1892 ret = fpga_dma_desc_td_fill(op, &desc->dec_req, input, output,
1893 in_length, k, in_offset, out_offset,
1894 ring_offset, cbs_in_op);
1895 if (unlikely(ret < 0))
1898 /* Update lengths */
1899 ret = rte_pktmbuf_trim(op->turbo_dec.hard_output.data,
1900 (crc24_overlap >> 3));
1901 #ifdef RTE_LIBRTE_BBDEV_DEBUG
1904 "The length to remove is greater than the length of the last segment");
1908 op->turbo_dec.hard_output.length += out_length;
1909 mbuf_total_left -= in_length;
1911 /* Update offsets */
1912 if (seg_total_left == in_length) {
1913 /* Go to the next mbuf */
1914 input = input->next;
1915 output = output->next;
1919 in_offset += in_length;
1920 out_offset += out_length;
1925 current_enqueued_cbs++;
1928 if (mbuf_total_left > 0) {
1930 "Some date still left for processing: mbuf_total_left = %u",
1935 return current_enqueued_cbs;
1939 fpga_enqueue_enc(struct rte_bbdev_queue_data *q_data,
1940 struct rte_bbdev_enc_op **ops, uint16_t num)
1943 uint16_t i, total_enqueued_cbs = 0;
1946 struct fpga_queue *q = q_data->queue_private;
1947 union fpga_dma_desc *desc;
1949 /* Check if queue is not full */
1950 if (unlikely(((q->tail + 1) & q->sw_ring_wrap_mask) ==
1954 /* Calculates available space */
1955 avail = (q->head_free_desc > q->tail) ?
1956 q->head_free_desc - q->tail - 1 :
1957 q->ring_ctrl_reg.ring_size + q->head_free_desc - q->tail - 1;
1959 for (i = 0; i < num; ++i) {
1960 if (ops[i]->turbo_enc.code_block_mode == 0) {
1961 cbs_in_op = get_num_cbs_in_op_enc(&ops[i]->turbo_enc);
1962 /* Check if there is available space for further
1965 if (unlikely(avail - cbs_in_op < 0))
1968 enqueued_cbs = enqueue_enc_one_op_tb(q, ops[i],
1969 total_enqueued_cbs, cbs_in_op);
1971 /* Check if there is available space for further
1974 if (unlikely(avail - 1 < 0))
1977 enqueued_cbs = enqueue_enc_one_op_cb(q, ops[i],
1978 total_enqueued_cbs);
1981 if (enqueued_cbs < 0)
1984 total_enqueued_cbs += enqueued_cbs;
1986 rte_bbdev_log_debug("enqueuing enc ops [%d/%d] | head %d | tail %d",
1987 total_enqueued_cbs, num,
1988 q->head_free_desc, q->tail);
1991 /* Set interrupt bit for last CB in enqueued ops. FPGA issues interrupt
1992 * only when all previous CBs were already processed.
1994 desc = q->ring_addr + ((q->tail + total_enqueued_cbs - 1)
1995 & q->sw_ring_wrap_mask);
1996 desc->enc_req.irq_en = q->irq_enable;
1998 fpga_dma_enqueue(q, total_enqueued_cbs, &q_data->queue_stats);
2001 q_data->queue_stats.enqueued_count += i;
2002 q_data->queue_stats.enqueue_err_count += num - i;
2008 fpga_enqueue_dec(struct rte_bbdev_queue_data *q_data,
2009 struct rte_bbdev_dec_op **ops, uint16_t num)
2012 uint16_t i, total_enqueued_cbs = 0;
2015 struct fpga_queue *q = q_data->queue_private;
2016 union fpga_dma_desc *desc;
2018 /* Check if queue is not full */
2019 if (unlikely(((q->tail + 1) & q->sw_ring_wrap_mask) ==
2023 /* Calculates available space */
2024 avail = (q->head_free_desc > q->tail) ?
2025 q->head_free_desc - q->tail - 1 :
2026 q->ring_ctrl_reg.ring_size + q->head_free_desc - q->tail - 1;
2028 for (i = 0; i < num; ++i) {
2029 if (ops[i]->turbo_dec.code_block_mode == 0) {
2030 cbs_in_op = get_num_cbs_in_op_dec(&ops[i]->turbo_dec);
2031 /* Check if there is available space for further
2034 if (unlikely(avail - cbs_in_op < 0))
2037 enqueued_cbs = enqueue_dec_one_op_tb(q, ops[i],
2038 total_enqueued_cbs, cbs_in_op);
2040 /* Check if there is available space for further
2043 if (unlikely(avail - 1 < 0))
2046 enqueued_cbs = enqueue_dec_one_op_cb(q, ops[i],
2047 total_enqueued_cbs);
2050 if (enqueued_cbs < 0)
2053 total_enqueued_cbs += enqueued_cbs;
2055 rte_bbdev_log_debug("enqueuing dec ops [%d/%d] | head %d | tail %d",
2056 total_enqueued_cbs, num,
2057 q->head_free_desc, q->tail);
2060 /* Set interrupt bit for last CB in enqueued ops. FPGA issues interrupt
2061 * only when all previous CBs were already processed.
2063 desc = q->ring_addr + ((q->tail + total_enqueued_cbs - 1)
2064 & q->sw_ring_wrap_mask);
2065 desc->dec_req.irq_en = q->irq_enable;
2067 fpga_dma_enqueue(q, total_enqueued_cbs, &q_data->queue_stats);
2070 q_data->queue_stats.enqueued_count += i;
2071 q_data->queue_stats.enqueue_err_count += num - i;
2077 dequeue_enc_one_op_cb(struct fpga_queue *q, struct rte_bbdev_enc_op **op,
2078 uint16_t desc_offset)
2080 union fpga_dma_desc *desc;
2083 /* Set current desc */
2084 desc = q->ring_addr + ((q->head_free_desc + desc_offset)
2085 & q->sw_ring_wrap_mask);
2088 if (desc->enc_req.done == 0)
2091 /* make sure the response is read atomically */
2094 rte_bbdev_log_debug("DMA response desc %p", desc);
2096 *op = desc->enc_req.op_addr;
2097 /* Check the decriptor error field, return 1 on error */
2098 desc_error = check_desc_error(desc->enc_req.error);
2099 (*op)->status = desc_error << RTE_BBDEV_DATA_ERROR;
2105 dequeue_enc_one_op_tb(struct fpga_queue *q, struct rte_bbdev_enc_op **op,
2106 uint16_t desc_offset)
2108 union fpga_dma_desc *desc;
2109 uint8_t cbs_in_op, cb_idx;
2113 /* Set descriptor */
2114 desc = q->ring_addr + ((q->head_free_desc + desc_offset)
2115 & q->sw_ring_wrap_mask);
2117 /* Verify if done bit is set */
2118 if (desc->enc_req.done == 0)
2121 /* Make sure the response is read atomically */
2124 /* Verify if done bit in all CBs is set */
2125 cbs_in_op = desc->enc_req.cbs_in_op;
2126 for (cb_idx = 1; cb_idx < cbs_in_op; ++cb_idx) {
2127 desc = q->ring_addr + ((q->head_free_desc + desc_offset +
2128 cb_idx) & q->sw_ring_wrap_mask);
2129 if (desc->enc_req.done == 0)
2133 /* Make sure the response is read atomically */
2136 for (cb_idx = 0; cb_idx < cbs_in_op; ++cb_idx) {
2137 desc = q->ring_addr + ((q->head_free_desc + desc_offset +
2138 cb_idx) & q->sw_ring_wrap_mask);
2139 /* Check the decriptor error field, return 1 on error */
2140 desc_error = check_desc_error(desc->enc_req.error);
2141 status |= desc_error << RTE_BBDEV_DATA_ERROR;
2142 rte_bbdev_log_debug("DMA response desc %p", desc);
2145 *op = desc->enc_req.op_addr;
2146 (*op)->status = status;
2151 dequeue_dec_one_op_cb(struct fpga_queue *q, struct rte_bbdev_dec_op **op,
2152 uint16_t desc_offset)
2154 union fpga_dma_desc *desc;
2156 /* Set descriptor */
2157 desc = q->ring_addr + ((q->head_free_desc + desc_offset)
2158 & q->sw_ring_wrap_mask);
2160 /* Verify done bit is set */
2161 if (desc->dec_req.done == 0)
2164 /* make sure the response is read atomically */
2167 #ifdef RTE_LIBRTE_BBDEV_DEBUG
2168 print_dma_dec_desc_debug_info(desc);
2172 *op = desc->dec_req.op_addr;
2173 /* FPGA reports in half-iterations, from 0 to 31. get ceiling */
2174 (*op)->turbo_dec.iter_count = (desc->dec_req.iter + 2) >> 1;
2175 /* crc_pass = 0 when decoder fails */
2176 (*op)->status = !(desc->dec_req.crc_pass) << RTE_BBDEV_CRC_ERROR;
2177 /* Check the decriptor error field, return 1 on error */
2178 desc_error = check_desc_error(desc->enc_req.error);
2179 (*op)->status |= desc_error << RTE_BBDEV_DATA_ERROR;
2184 dequeue_dec_one_op_tb(struct fpga_queue *q, struct rte_bbdev_dec_op **op,
2185 uint16_t desc_offset)
2187 union fpga_dma_desc *desc;
2188 uint8_t cbs_in_op, cb_idx, iter_count = 0;
2191 /* Set descriptor */
2192 desc = q->ring_addr + ((q->head_free_desc + desc_offset)
2193 & q->sw_ring_wrap_mask);
2195 /* Verify if done bit is set */
2196 if (desc->dec_req.done == 0)
2199 /* Make sure the response is read atomically */
2202 /* Verify if done bit in all CBs is set */
2203 cbs_in_op = desc->dec_req.cbs_in_op;
2204 for (cb_idx = 1; cb_idx < cbs_in_op; ++cb_idx) {
2205 desc = q->ring_addr + ((q->head_free_desc + desc_offset +
2206 cb_idx) & q->sw_ring_wrap_mask);
2207 if (desc->dec_req.done == 0)
2211 /* Make sure the response is read atomically */
2214 for (cb_idx = 0; cb_idx < cbs_in_op; ++cb_idx) {
2215 desc = q->ring_addr + ((q->head_free_desc + desc_offset +
2216 cb_idx) & q->sw_ring_wrap_mask);
2217 /* get max iter_count for all CBs in op */
2218 iter_count = RTE_MAX(iter_count, (uint8_t) desc->dec_req.iter);
2219 /* crc_pass = 0 when decoder fails, one fails all */
2220 status |= !(desc->dec_req.crc_pass) << RTE_BBDEV_CRC_ERROR;
2221 /* Check the decriptor error field, return 1 on error */
2222 desc_error = check_desc_error(desc->enc_req.error);
2223 status |= desc_error << RTE_BBDEV_DATA_ERROR;
2224 rte_bbdev_log_debug("DMA response desc %p", desc);
2227 *op = desc->dec_req.op_addr;
2229 /* FPGA reports in half-iterations, get ceiling */
2230 (*op)->turbo_dec.iter_count = (iter_count + 2) >> 1;
2231 (*op)->status = status;
2236 fpga_dequeue_enc(struct rte_bbdev_queue_data *q_data,
2237 struct rte_bbdev_enc_op **ops, uint16_t num)
2239 struct fpga_queue *q = q_data->queue_private;
2240 uint32_t avail = (q->tail - q->head_free_desc) & q->sw_ring_wrap_mask;
2242 uint16_t dequeued_cbs = 0;
2243 struct rte_bbdev_enc_op *op;
2246 for (i = 0; (i < num) && (dequeued_cbs < avail); ++i) {
2247 op = (q->ring_addr + ((q->head_free_desc + dequeued_cbs)
2248 & q->sw_ring_wrap_mask))->enc_req.op_addr;
2249 if (op->turbo_enc.code_block_mode == 0)
2250 ret = dequeue_enc_one_op_tb(q, &ops[i], dequeued_cbs);
2252 ret = dequeue_enc_one_op_cb(q, &ops[i], dequeued_cbs);
2257 dequeued_cbs += ret;
2259 rte_bbdev_log_debug("dequeuing enc ops [%d/%d] | head %d | tail %d",
2260 dequeued_cbs, num, q->head_free_desc, q->tail);
2264 q->head_free_desc = (q->head_free_desc + dequeued_cbs) &
2265 q->sw_ring_wrap_mask;
2268 q_data->queue_stats.dequeued_count += i;
2274 fpga_dequeue_dec(struct rte_bbdev_queue_data *q_data,
2275 struct rte_bbdev_dec_op **ops, uint16_t num)
2277 struct fpga_queue *q = q_data->queue_private;
2278 uint32_t avail = (q->tail - q->head_free_desc) & q->sw_ring_wrap_mask;
2280 uint16_t dequeued_cbs = 0;
2281 struct rte_bbdev_dec_op *op;
2284 for (i = 0; (i < num) && (dequeued_cbs < avail); ++i) {
2285 op = (q->ring_addr + ((q->head_free_desc + dequeued_cbs)
2286 & q->sw_ring_wrap_mask))->dec_req.op_addr;
2287 if (op->turbo_dec.code_block_mode == 0)
2288 ret = dequeue_dec_one_op_tb(q, &ops[i], dequeued_cbs);
2290 ret = dequeue_dec_one_op_cb(q, &ops[i], dequeued_cbs);
2295 dequeued_cbs += ret;
2297 rte_bbdev_log_debug("dequeuing dec ops [%d/%d] | head %d | tail %d",
2298 dequeued_cbs, num, q->head_free_desc, q->tail);
2302 q->head_free_desc = (q->head_free_desc + dequeued_cbs) &
2303 q->sw_ring_wrap_mask;
2306 q_data->queue_stats.dequeued_count += i;
2311 /* Initialization Function */
2313 fpga_lte_fec_init(struct rte_bbdev *dev, struct rte_pci_driver *drv)
2315 struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
2317 dev->dev_ops = &fpga_ops;
2318 dev->enqueue_enc_ops = fpga_enqueue_enc;
2319 dev->enqueue_dec_ops = fpga_enqueue_dec;
2320 dev->dequeue_enc_ops = fpga_dequeue_enc;
2321 dev->dequeue_dec_ops = fpga_dequeue_dec;
2323 ((struct fpga_lte_fec_device *) dev->data->dev_private)->pf_device =
2324 !strcmp(drv->driver.name,
2325 RTE_STR(FPGA_LTE_FEC_PF_DRIVER_NAME));
2326 ((struct fpga_lte_fec_device *) dev->data->dev_private)->mmio_base =
2327 pci_dev->mem_resource[0].addr;
2329 rte_bbdev_log_debug(
2330 "Init device %s [%s] @ virtaddr %p phyaddr %#"PRIx64,
2331 drv->driver.name, dev->data->name,
2332 (void *)pci_dev->mem_resource[0].addr,
2333 pci_dev->mem_resource[0].phys_addr);
2337 fpga_lte_fec_probe(struct rte_pci_driver *pci_drv,
2338 struct rte_pci_device *pci_dev)
2340 struct rte_bbdev *bbdev = NULL;
2341 char dev_name[RTE_BBDEV_NAME_MAX_LEN];
2343 if (pci_dev == NULL) {
2344 rte_bbdev_log(ERR, "NULL PCI device");
2348 rte_pci_device_name(&pci_dev->addr, dev_name, sizeof(dev_name));
2350 /* Allocate memory to be used privately by drivers */
2351 bbdev = rte_bbdev_allocate(pci_dev->device.name);
2355 /* allocate device private memory */
2356 bbdev->data->dev_private = rte_zmalloc_socket(dev_name,
2357 sizeof(struct fpga_lte_fec_device), RTE_CACHE_LINE_SIZE,
2358 pci_dev->device.numa_node);
2360 if (bbdev->data->dev_private == NULL) {
2362 "Allocate of %zu bytes for device \"%s\" failed",
2363 sizeof(struct fpga_lte_fec_device), dev_name);
2364 rte_bbdev_release(bbdev);
2368 /* Fill HW specific part of device structure */
2369 bbdev->device = &pci_dev->device;
2370 bbdev->intr_handle = &pci_dev->intr_handle;
2371 bbdev->data->socket_id = pci_dev->device.numa_node;
2373 /* Invoke FEC FPGA device initialization function */
2374 fpga_lte_fec_init(bbdev, pci_drv);
2376 rte_bbdev_log_debug("bbdev id = %u [%s]",
2377 bbdev->data->dev_id, dev_name);
2379 struct fpga_lte_fec_device *d = bbdev->data->dev_private;
2380 uint32_t version_id = fpga_reg_read_32(d->mmio_base,
2381 FPGA_LTE_FEC_VERSION_ID);
2382 rte_bbdev_log(INFO, "FEC FPGA RTL v%u.%u",
2383 ((uint16_t)(version_id >> 16)), ((uint16_t)version_id));
2385 #ifdef RTE_LIBRTE_BBDEV_DEBUG
2386 if (!strcmp(pci_drv->driver.name,
2387 RTE_STR(FPGA_LTE_FEC_PF_DRIVER_NAME)))
2388 print_static_reg_debug_info(d->mmio_base);
2394 fpga_lte_fec_remove(struct rte_pci_device *pci_dev)
2396 struct rte_bbdev *bbdev;
2400 if (pci_dev == NULL)
2404 bbdev = rte_bbdev_get_named_dev(pci_dev->device.name);
2405 if (bbdev == NULL) {
2407 "Couldn't find HW dev \"%s\" to uninitialise it",
2408 pci_dev->device.name);
2411 dev_id = bbdev->data->dev_id;
2413 /* free device private memory before close */
2414 rte_free(bbdev->data->dev_private);
2417 ret = rte_bbdev_close(dev_id);
2420 "Device %i failed to close during uninit: %i",
2423 /* release bbdev from library */
2424 ret = rte_bbdev_release(bbdev);
2426 rte_bbdev_log(ERR, "Device %i failed to uninit: %i", dev_id,
2429 rte_bbdev_log_debug("Destroyed bbdev = %u", dev_id);
2435 set_default_fpga_conf(struct rte_fpga_lte_fec_conf *def_conf)
2437 /* clear default configuration before initialization */
2438 memset(def_conf, 0, sizeof(struct rte_fpga_lte_fec_conf));
2439 /* Set pf mode to true */
2440 def_conf->pf_mode_en = true;
2442 /* Set ratio between UL and DL to 1:1 (unit of weight is 3 CBs) */
2443 def_conf->ul_bandwidth = 3;
2444 def_conf->dl_bandwidth = 3;
2446 /* Set Load Balance Factor to 64 */
2447 def_conf->dl_load_balance = 64;
2448 def_conf->ul_load_balance = 64;
2451 /* Initial configuration of FPGA LTE FEC device */
2453 rte_fpga_lte_fec_configure(const char *dev_name,
2454 const struct rte_fpga_lte_fec_conf *conf)
2456 uint32_t payload_32, address;
2457 uint16_t payload_16;
2459 uint16_t q_id, vf_id, total_q_id, total_ul_q_id, total_dl_q_id;
2460 struct rte_bbdev *bbdev = rte_bbdev_get_named_dev(dev_name);
2461 struct rte_fpga_lte_fec_conf def_conf;
2463 if (bbdev == NULL) {
2465 "Invalid dev_name (%s), or device is not yet initialised",
2470 struct fpga_lte_fec_device *d = bbdev->data->dev_private;
2474 "FPGA Configuration was not provided. Default configuration will be loaded.");
2475 set_default_fpga_conf(&def_conf);
2480 * Configure UL:DL ratio.
2484 payload_16 = (conf->dl_bandwidth << 8) | conf->ul_bandwidth;
2485 address = FPGA_LTE_FEC_CONFIGURATION;
2486 fpga_reg_write_16(d->mmio_base, address, payload_16);
2488 /* Clear all queues registers */
2489 payload_32 = FPGA_INVALID_HW_QUEUE_ID;
2490 for (q_id = 0; q_id < FPGA_TOTAL_NUM_QUEUES; ++q_id) {
2491 address = (q_id << 2) + FPGA_LTE_FEC_QUEUE_MAP;
2492 fpga_reg_write_32(d->mmio_base, address, payload_32);
2496 * If PF mode is enabled allocate all queues for PF only.
2498 * For VF mode each VF can have different number of UL and DL queues.
2499 * Total number of queues to configure cannot exceed FPGA
2500 * capabilities - 64 queues - 32 queues for UL and 32 queues for DL.
2501 * Queues mapping is done according to configuration:
2507 * | conf->vf_dl_queues_number[0] - 1 | 0 |
2508 * | conf->vf_dl_queues_number[0] | 1 |
2510 * | conf->vf_dl_queues_number[1] - 1 | 1 |
2512 * | conf->vf_dl_queues_number[7] - 1 | 7 |
2518 * | conf->vf_ul_queues_number[0] - 1 | 0 |
2519 * | conf->vf_ul_queues_number[0] | 1 |
2521 * | conf->vf_ul_queues_number[1] - 1 | 1 |
2523 * | conf->vf_ul_queues_number[7] - 1 | 7 |
2525 * Example of configuration:
2526 * conf->vf_ul_queues_number[0] = 4; -> 4 UL queues for VF0
2527 * conf->vf_dl_queues_number[0] = 4; -> 4 DL queues for VF0
2528 * conf->vf_ul_queues_number[1] = 2; -> 2 UL queues for VF1
2529 * conf->vf_dl_queues_number[1] = 2; -> 2 DL queues for VF1
2549 if (conf->pf_mode_en) {
2551 for (q_id = 0; q_id < FPGA_TOTAL_NUM_QUEUES; ++q_id) {
2552 address = (q_id << 2) + FPGA_LTE_FEC_QUEUE_MAP;
2553 fpga_reg_write_32(d->mmio_base, address, payload_32);
2556 /* Calculate total number of UL and DL queues to configure */
2557 total_ul_q_id = total_dl_q_id = 0;
2558 for (vf_id = 0; vf_id < FPGA_LTE_FEC_NUM_VFS; ++vf_id) {
2559 total_ul_q_id += conf->vf_ul_queues_number[vf_id];
2560 total_dl_q_id += conf->vf_dl_queues_number[vf_id];
2562 total_q_id = total_dl_q_id + total_ul_q_id;
2564 * Check if total number of queues to configure does not exceed
2565 * FPGA capabilities (64 queues - 32 UL and 32 DL queues)
2567 if ((total_ul_q_id > FPGA_NUM_UL_QUEUES) ||
2568 (total_dl_q_id > FPGA_NUM_DL_QUEUES) ||
2569 (total_q_id > FPGA_TOTAL_NUM_QUEUES)) {
2571 "FPGA Configuration failed. Too many queues to configure: UL_Q %u, DL_Q %u, FPGA_Q %u",
2572 total_ul_q_id, total_dl_q_id,
2573 FPGA_TOTAL_NUM_QUEUES);
2577 for (vf_id = 0; vf_id < FPGA_LTE_FEC_NUM_VFS; ++vf_id) {
2578 for (q_id = 0; q_id < conf->vf_ul_queues_number[vf_id];
2579 ++q_id, ++total_ul_q_id) {
2580 address = (total_ul_q_id << 2) +
2581 FPGA_LTE_FEC_QUEUE_MAP;
2582 payload_32 = ((0x80 + vf_id) << 16) | 0x1;
2583 fpga_reg_write_32(d->mmio_base, address,
2588 for (vf_id = 0; vf_id < FPGA_LTE_FEC_NUM_VFS; ++vf_id) {
2589 for (q_id = 0; q_id < conf->vf_dl_queues_number[vf_id];
2590 ++q_id, ++total_dl_q_id) {
2591 address = ((total_dl_q_id + FPGA_NUM_UL_QUEUES)
2592 << 2) + FPGA_LTE_FEC_QUEUE_MAP;
2593 payload_32 = ((0x80 + vf_id) << 16) | 0x1;
2594 fpga_reg_write_32(d->mmio_base, address,
2600 /* Setting Load Balance Factor */
2601 payload_16 = (conf->dl_load_balance << 8) | (conf->ul_load_balance);
2602 address = FPGA_LTE_FEC_LOAD_BALANCE_FACTOR;
2603 fpga_reg_write_16(d->mmio_base, address, payload_16);
2605 /* Setting length of ring descriptor entry */
2606 payload_16 = FPGA_RING_DESC_ENTRY_LENGTH;
2607 address = FPGA_LTE_FEC_RING_DESC_LEN;
2608 fpga_reg_write_16(d->mmio_base, address, payload_16);
2610 /* Setting FLR timeout value */
2611 payload_16 = conf->flr_time_out;
2612 address = FPGA_LTE_FEC_FLR_TIME_OUT;
2613 fpga_reg_write_16(d->mmio_base, address, payload_16);
2615 /* Queue PF/VF mapping table is ready */
2617 address = FPGA_LTE_FEC_QUEUE_PF_VF_MAP_DONE;
2618 fpga_reg_write_8(d->mmio_base, address, payload_8);
2620 rte_bbdev_log_debug("PF FPGA LTE FEC configuration complete for %s",
2623 #ifdef RTE_LIBRTE_BBDEV_DEBUG
2624 print_static_reg_debug_info(d->mmio_base);
2629 /* FPGA LTE FEC PCI PF address map */
2630 static struct rte_pci_id pci_id_fpga_lte_fec_pf_map[] = {
2632 RTE_PCI_DEVICE(FPGA_LTE_FEC_VENDOR_ID,
2633 FPGA_LTE_FEC_PF_DEVICE_ID)
2638 static struct rte_pci_driver fpga_lte_fec_pci_pf_driver = {
2639 .probe = fpga_lte_fec_probe,
2640 .remove = fpga_lte_fec_remove,
2641 .id_table = pci_id_fpga_lte_fec_pf_map,
2642 .drv_flags = RTE_PCI_DRV_NEED_MAPPING
2645 /* FPGA LTE FEC PCI VF address map */
2646 static struct rte_pci_id pci_id_fpga_lte_fec_vf_map[] = {
2648 RTE_PCI_DEVICE(FPGA_LTE_FEC_VENDOR_ID,
2649 FPGA_LTE_FEC_VF_DEVICE_ID)
2654 static struct rte_pci_driver fpga_lte_fec_pci_vf_driver = {
2655 .probe = fpga_lte_fec_probe,
2656 .remove = fpga_lte_fec_remove,
2657 .id_table = pci_id_fpga_lte_fec_vf_map,
2658 .drv_flags = RTE_PCI_DRV_NEED_MAPPING
2662 RTE_PMD_REGISTER_PCI(FPGA_LTE_FEC_PF_DRIVER_NAME, fpga_lte_fec_pci_pf_driver);
2663 RTE_PMD_REGISTER_PCI_TABLE(FPGA_LTE_FEC_PF_DRIVER_NAME,
2664 pci_id_fpga_lte_fec_pf_map);
2665 RTE_PMD_REGISTER_PCI(FPGA_LTE_FEC_VF_DRIVER_NAME, fpga_lte_fec_pci_vf_driver);
2666 RTE_PMD_REGISTER_PCI_TABLE(FPGA_LTE_FEC_VF_DRIVER_NAME,
2667 pci_id_fpga_lte_fec_vf_map);