DPDK patches and discussions
 help / color / mirror / Atom feed
From: "Chautru, Nicolas" <nicolas.chautru@intel.com>
To: John Miller <john.miller@atomicrules.com>
Cc: "dev@dpdk.org" <dev@dpdk.org>,
	"ed.czeck@atomicrules.com" <ed.czeck@atomicrules.com>,
	Shepard Siegel <shepard.siegel@atomicrules.com>,
	Maxime Coquelin <maxime.coquelin@redhat.com>
Subject: RE: [PATCH 11/14] baseband/ark: introduce ark baseband driver custom functions
Date: Wed, 26 Oct 2022 23:22:56 +0000	[thread overview]
Message-ID: <BY5PR11MB445192CE9AB99004B33938BAF8309@BY5PR11MB4451.namprd11.prod.outlook.com> (raw)
In-Reply-To: <20221026194613.1008232-11-john.miller@atomicrules.com>

Hi John, 

> -----Original Message-----
> From: John Miller <john.miller@atomicrules.com>
> Sent: Wednesday, October 26, 2022 12:46 PM
> To: Chautru, Nicolas <nicolas.chautru@intel.com>
> Cc: dev@dpdk.org; ed.czeck@atomicrules.com; Shepard Siegel
> <shepard.siegel@atomicrules.com>; John Miller
> <john.miller@atomicrules.com>
> Subject: [PATCH 11/14] baseband/ark: introduce ark baseband driver custom
> functions
> 
> This patch introduces the Arkville baseband device driver custom functions.
> 
> Signed-off-by: John Miller <john.miller@atomicrules.com>
> ---
>  drivers/baseband/ark/ark_bbdev_custom.c | 201 ++++++++++++++++++++++++
> drivers/baseband/ark/ark_bbdev_custom.h |  30 ++++
>  2 files changed, 231 insertions(+)
>  create mode 100644 drivers/baseband/ark/ark_bbdev_custom.c
>  create mode 100644 drivers/baseband/ark/ark_bbdev_custom.h
> 
> diff --git a/drivers/baseband/ark/ark_bbdev_custom.c
> b/drivers/baseband/ark/ark_bbdev_custom.c
> new file mode 100644
> index 0000000000..6b1553abe1
> --- /dev/null
> +++ b/drivers/baseband/ark/ark_bbdev_custom.c
> @@ -0,0 +1,201 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2016-2021 Atomic Rules LLC  */
> +
> +#include <rte_bbdev.h>
> +#include <rte_bbdev_pmd.h>
> +
> +#include <rte_mbuf.h>
> +#include <rte_hexdump.h>	/* For debug */
> +
> +
> +#include "ark_bbdev_common.h"
> +#include "ark_bbdev_custom.h"
> +
> +/* It is expected that functions in this file will be modified based on
> + * specifics of the FPGA hardware beyond the core Arkville
> + * components.
> + */
> +
> +/* bytyes must be range of 0 to 20 */
> +static inline
> +uint8_t ark_bb_cvt_bytes_meta_cnt(size_t bytes) {
> +	return (bytes + 3) / 8;
> +}
> +
> +void
> +ark_bbdev_info_get(struct rte_bbdev *dev,
> +		   struct rte_bbdev_driver_info *dev_info) {
> +	struct ark_bbdevice *ark_bb =  dev->data->dev_private;
> +

In your documentation in first commit you mentioned this
* Support for LDPC encode and decode operations.
* Support for Turbo encode and decode operations.
But I only see LDPC below. More generally not really matching the doc I think. Good to have the code and docs in same commits for that reason. 

> +	static const struct rte_bbdev_op_cap bbdev_capabilities[] = {
> +		{
> +			.type = RTE_BBDEV_OP_LDPC_DEC,
> +			.cap.ldpc_dec = {
> +				.capability_flags =
> +					RTE_BBDEV_LDPC_CRC_24B_ATTACH
> |
> +					RTE_BBDEV_LDPC_RATE_MATCH,

It doesn't look right
Basically there se flags are not for LDPC_DEC but for the encoder
There is no HARQ combine etc?
I would have expected scatter gather here as well based on your documentation

> +				.num_buffers_src =
> +
> 	RTE_BBDEV_LDPC_MAX_CODE_BLOCKS,
> +				.num_buffers_hard_out =
> +
> 	RTE_BBDEV_LDPC_MAX_CODE_BLOCKS
> +			}
> +		},
> +		{
> +			.type = RTE_BBDEV_OP_LDPC_ENC,
> +			.cap.ldpc_enc = {
> +				.capability_flags =
> +					RTE_BBDEV_LDPC_CRC_24B_ATTACH
> |
> +					RTE_BBDEV_LDPC_RATE_MATCH,
> +				.num_buffers_src =
> +
> 	RTE_BBDEV_LDPC_MAX_CODE_BLOCKS,
> +				.num_buffers_dst =
> +
> 	RTE_BBDEV_LDPC_MAX_CODE_BLOCKS
> +			}
> +		},
> +		RTE_BBDEV_END_OF_CAPABILITIES_LIST(),
> +	};
> +
> +	static struct rte_bbdev_queue_conf default_queue_conf = {
> +		.queue_size = RTE_BBDEV_QUEUE_SIZE_LIMIT,
> +	};
> +
> +	default_queue_conf.socket = dev->data->socket_id;
> +
> +	dev_info->driver_name = RTE_STR(DRIVER_NAME);
> +	dev_info->max_num_queues = ark_bb->max_nb_queues;
> +	dev_info->queue_size_lim = RTE_BBDEV_QUEUE_SIZE_LIMIT;
> +	dev_info->hardware_accelerated = true;
> +	dev_info->max_dl_queue_priority = 0;
> +	dev_info->max_ul_queue_priority = 0;
> +	dev_info->default_queue_conf = default_queue_conf;
> +	dev_info->capabilities = bbdev_capabilities;
> +	dev_info->cpu_flag_reqs = NULL;
> +	dev_info->min_alignment = 4;

Is there really a 4 Bytes alignment requirement per code blocks? That sounds extremely cumbersome, is that what you really mean?

> +
> +}
> +
> +/* Structure defining layout of the ldpc command struct */ struct
> +ark_bb_ldpc_enc_meta {
> +	uint16_t header;
> +	uint8_t rv_index:2,
> +		basegraph:1,
> +		code_block_mode:1,
> +		rfu_71_68:4;

What is this? Just curious. 

> +
> +	uint8_t q_m;
> +	uint32_t e_ea;
> +	uint32_t eb;
> +	uint8_t c;
> +	uint8_t cab;
> +	uint16_t n_cb;
> +	uint16_t pad;
> +	uint16_t trailer;
> +} __rte_packed;
> +
> +/* The size must be less then 20 Bytes */ static_assert(sizeof(struct
> +ark_bb_ldpc_enc_meta) <= 20, "struct size");
> +
> +/* Custom operation on equeue ldpc operation  */

Typo enqueue

> +/* Do these function need queue number? */

Who is the question for? Through bbdev api the queue index is expected, and from your documentation I believe you support multiple queues. 

> +/* Maximum of 20 bytes */
> +int
> +ark_bb_user_enqueue_ldpc_enc(struct rte_bbdev_enc_op *enc_op,
> +			  uint32_t *meta, uint8_t *meta_cnt) {
> +	struct rte_bbdev_op_ldpc_enc *ldpc_enc_op = &enc_op->ldpc_enc;
> +	struct ark_bb_ldpc_enc_meta *src = (struct ark_bb_ldpc_enc_meta
> +*)meta;
> +
> +	src->header = 0x4321;	/* For testings */
> +	src->trailer = 0xFEDC;
> +
> +	src->rv_index = ldpc_enc_op->rv_index;
> +	src->basegraph = ldpc_enc_op->basegraph;
> +	src->code_block_mode = ldpc_enc_op->code_block_mode;
> +
> +	src->q_m = ldpc_enc_op->q_m;
> +	src->e_ea = 0xABCD;
> +	src->eb = ldpc_enc_op->tb_params.eb;
> +	src->c = ldpc_enc_op->tb_params.c;
> +	src->cab = ldpc_enc_op->tb_params.cab;
> +
> +	src->n_cb = 0;
> +
> +	meta[0] = 0x11111110;
> +	meta[1] = 0x22222220;
> +	meta[2] = 0x33333330;
> +	meta[3] = 0x44444440;
> +	meta[4] = 0x55555550;

Should these be defined separately?

> +
> +	*meta_cnt = ark_bb_cvt_bytes_meta_cnt(
> +			sizeof(struct ark_bb_ldpc_enc_meta));
> +	return 0;
> +}
> +
> +/* Custom operation on dequeue ldpc operation  */ int
> +ark_bb_user_dequeue_ldpc_enc(struct rte_bbdev_enc_op *enc_op,
> +			     const uint32_t *usermeta)
> +{
> +	static int dump;	/* = 0 */
> +	/* Just compare with what was sent? */
> +	uint32_t meta_in[5] = {0};
> +	uint8_t  meta_cnt;
> +
> +	ark_bb_user_enqueue_ldpc_enc(enc_op, meta_in, &meta_cnt);
> +	if (memcmp(usermeta, meta_in, 3 + (meta_cnt * 8))) {
> +		fprintf(stderr,
> +			"------------------------------------------\n");
> +		rte_hexdump(stdout, "meta difference for lpdc_enc IN",
> +			    meta_in, 20);
> +		rte_hexdump(stdout, "meta difference for lpdc_enc OUT",
> +			    usermeta, 20);
> +	} else if (dump) {
> +		rte_hexdump(stdout, "DUMP lpdc_enc IN", usermeta, 20);
> +		dump--;
> +	}
> +
> +	return 0;
> +}
> +
> +
> +/* Turbo op call backs for user meta data */ int
> +ark_bb_user_enqueue_ldpc_dec(struct rte_bbdev_dec_op *enc_op,
> +				 uint32_t *meta, uint8_t *meta_cnt) {
> +	RTE_SET_USED(enc_op);

Is the implementation missing?

The enc_op should be called differently. 

> +	meta[0] = 0xF1111110;
> +	meta[1] = 0xF2222220;
> +	meta[2] = 0xF3333330;
> +	meta[3] = 0xF4444440;
> +	meta[4] = 0xF5555550;
> +
> +	*meta_cnt = ark_bb_cvt_bytes_meta_cnt(20);
> +	return 0;
> +}
> +
> +int ark_bb_user_dequeue_ldpc_dec(struct rte_bbdev_dec_op *enc_op,
> +				 const uint32_t *usermeta)
> +{
> +	RTE_SET_USED(enc_op);
> +	static int dump;	/* = 0 */
> +	/* Just compare with what was sent? */

That looks like still testcode isn't it? Doing some loopback. 

> +	uint32_t meta_in[5] = {0};
> +	uint8_t  meta_cnt;
> +
> +	ark_bb_user_enqueue_ldpc_dec(enc_op, meta_in, &meta_cnt);
> +	if (memcmp(usermeta, meta_in, 3 + (meta_cnt * 8))) {
> +		fprintf(stderr,
> +			"------------------------------------------\n");
> +		rte_hexdump(stdout, "meta difference for lpdc_enc IN",
> +			    meta_in, 20);
> +		rte_hexdump(stdout, "meta difference for lpdc_enc OUT",
> +			    usermeta, 20);
> +	} else if (dump) {
> +		rte_hexdump(stdout, "DUMP lpdc_enc IN", usermeta, 20);
> +		dump--;
> +	}
> +	return 0;
> +}
> diff --git a/drivers/baseband/ark/ark_bbdev_custom.h
> b/drivers/baseband/ark/ark_bbdev_custom.h
> new file mode 100644
> index 0000000000..32a2ef6bb6
> --- /dev/null
> +++ b/drivers/baseband/ark/ark_bbdev_custom.h
> @@ -0,0 +1,30 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2016-2021 Atomic Rules LLC  */
> +
> +#ifndef _ARK_BBDEV_CUSTOM_H_
> +#define _ARK_BBDEV_CUSTOM_H_
> +
> +#include <stdint.h>
> +
> +/* Forward declarations */
> +struct rte_bbdev;
> +struct rte_bbdev_driver_info;
> +struct rte_bbdev_enc_op;
> +struct rte_bbdev_dec_op;
> +struct rte_mbuf;
> +
> +void ark_bbdev_info_get(struct rte_bbdev *dev,
> +			struct rte_bbdev_driver_info *dev_info);
> +
> +int ark_bb_user_enqueue_ldpc_dec(struct rte_bbdev_dec_op *enc_op,
> +				 uint32_t *meta, uint8_t *meta_cnt); int
> +ark_bb_user_dequeue_ldpc_dec(struct rte_bbdev_dec_op *enc_op,
> +				 const uint32_t *usermeta);
> +
> +int ark_bb_user_enqueue_ldpc_enc(struct rte_bbdev_enc_op *enc_op,
> +				 uint32_t *meta, uint8_t *meta_cnt); int
> +ark_bb_user_dequeue_ldpc_enc(struct rte_bbdev_enc_op *enc_op,
> +				 const uint32_t *usermeta);
> +
> +#endif
> --
> 2.25.1


  reply	other threads:[~2022-10-26 23:23 UTC|newest]

Thread overview: 20+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-10-26 19:46 [PATCH 01/14] doc/guides/bbdevs: add ark baseband device documentation John Miller
2022-10-26 19:46 ` [PATCH 02/14] common/ark: create common subdirectory for baseband support John Miller
2022-10-26 19:46 ` [PATCH 03/14] common/ark: move common files to common subdirectory John Miller
2022-10-26 19:46 ` [PATCH 04/14] common/meson.build: John Miller
2022-10-26 19:46 ` [PATCH 05/14] net/ark: remove build files moved to common John Miller
2022-10-26 19:46 ` [PATCH 06/14] common/ark: update version map file John Miller
2022-10-26 19:46 ` [PATCH 07/14] common/ark: avoid exporting internal functions John Miller
2022-10-26 19:46 ` [PATCH 08/14] net/ark: add ark PMD log interface John Miller
2022-10-26 19:46 ` [PATCH 09/14] common/ark: add VF support to caps record John Miller
2022-10-26 19:46 ` [PATCH 10/14] baseband/ark: introduce ark baseband driver John Miller
2022-10-26 23:11   ` Chautru, Nicolas
2022-10-31 17:33     ` John Miller
2022-10-31 21:15       ` Chautru, Nicolas
2022-10-26 19:46 ` [PATCH 11/14] baseband/ark: introduce ark baseband driver custom functions John Miller
2022-10-26 23:22   ` Chautru, Nicolas [this message]
2022-11-04 11:35     ` John Miller
2022-10-26 19:46 ` [PATCH 12/14] baseband/ark: introduce ark baseband driver common functions John Miller
2022-10-26 19:46 ` [PATCH 13/14] baseband/ark: introduce ark baseband build files John Miller
2022-10-26 19:46 ` [PATCH 14/14] baseband/meson.build: John Miller
2022-12-15 14:18 ` [PATCH 01/14] doc/guides/bbdevs: add ark baseband device documentation Maxime Coquelin

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=BY5PR11MB445192CE9AB99004B33938BAF8309@BY5PR11MB4451.namprd11.prod.outlook.com \
    --to=nicolas.chautru@intel.com \
    --cc=dev@dpdk.org \
    --cc=ed.czeck@atomicrules.com \
    --cc=john.miller@atomicrules.com \
    --cc=maxime.coquelin@redhat.com \
    --cc=shepard.siegel@atomicrules.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).