[PATCH v3 0/7] baseband/acc: changes for 23.03

[PATCH v3 0/7] baseband/acc: changes for 23.03

[Hernan Vargas]

v3: Remove unused mempool header.
v2: Reordered commits and added one more fix.
v1: Upstreaming ACC100 changes for 23.03.

Hernan Vargas (7):
  baseband/acc: acc100 free harq layout pointer
  baseband/acc: acc100 fix iteration counter in TB
  baseband/acc: acc100 fix multiplexing multiple ops
  baseband/acc: acc100 fix queue mapping to 64 bits
  baseband/acc: acc100 use define constant
  baseband/acc: acc100 use desc helper functions
  baseband/acc: acc100 ignore missing mempools

 drivers/baseband/acc/rte_acc100_pmd.c | 65 +++++++++++----------------
 1 file changed, 25 insertions(+), 40 deletions(-)

-- 
2.37.1

[PATCH v3 1/7] baseband/acc: acc100 free harq layout pointer

[Hernan Vargas]

Explicitly call rte_free for harq_layout pointer in dev_close function
to prevent memory leak.

Fixes: ba2262fe16c ("baseband/acc100: fix close cleanup")
Cc: stable@dpdk.org

Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
---
 drivers/baseband/acc/rte_acc100_pmd.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
index b3164529dd..146a22bd64 100644
--- a/drivers/baseband/acc/rte_acc100_pmd.c
+++ b/drivers/baseband/acc/rte_acc100_pmd.c
@@ -620,6 +620,7 @@ acc100_dev_close(struct rte_bbdev *dev)
 		rte_free(d->tail_ptrs);
 		rte_free(d->info_ring);
 		rte_free(d->sw_rings_base);
+		rte_free(d->harq_layout);
 		d->sw_rings_base = NULL;
 		d->tail_ptrs = NULL;
 		d->info_ring = NULL;
-- 
2.37.1

[PATCH v3 2/7] baseband/acc: acc100 fix iteration counter in TB

[Hernan Vargas]

Use ldpc or turbo iteration counter based on the operation type.

Fixes: 5ad5060f8f7 ("baseband/acc100: add LDPC processing functions")
Cc: stable@dpdk.org

Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
---
 drivers/baseband/acc/rte_acc100_pmd.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
index 146a22bd64..a00314c957 100644
--- a/drivers/baseband/acc/rte_acc100_pmd.c
+++ b/drivers/baseband/acc/rte_acc100_pmd.c
@@ -4033,8 +4033,12 @@ dequeue_dec_one_op_tb(struct acc_queue *q, struct rte_bbdev_dec_op **ref_op,
 		/* CRC invalid if error exists */
 		if (!op->status)
 			op->status |= rsp.crc_status << RTE_BBDEV_CRC_ERROR;
-		op->turbo_dec.iter_count = RTE_MAX((uint8_t) rsp.iter_cnt,
-				op->turbo_dec.iter_count);
+		if (q->op_type == RTE_BBDEV_OP_LDPC_DEC)
+			op->ldpc_dec.iter_count = RTE_MAX((uint8_t) rsp.iter_cnt,
+					op->ldpc_dec.iter_count);
+		else
+			op->turbo_dec.iter_count = RTE_MAX((uint8_t) rsp.iter_cnt,
+					op->turbo_dec.iter_count);
 
 		/* Check if this is the last desc in batch (Atomic Queue) */
 		if (desc->req.last_desc_in_batch) {
-- 
2.37.1

[PATCH v3 3/7] baseband/acc: acc100 fix multiplexing multiple ops

[Hernan Vargas]

Function to check if multiplexing is possible for multiple operations in
one descriptor returns number of enqueued ops.

Fixes: 32e8b7ea35d ("baseband/acc100: refactor to segregate common code")

Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
---
 drivers/baseband/acc/rte_acc100_pmd.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
index a00314c957..eb18cc2215 100644
--- a/drivers/baseband/acc/rte_acc100_pmd.c
+++ b/drivers/baseband/acc/rte_acc100_pmd.c
@@ -3421,9 +3421,9 @@ acc100_enqueue_ldpc_enc_cb(struct rte_bbdev_queue_data *q_data,
 		}
 		avail--;
 		enq = RTE_MIN(left, ACC_MUX_5GDL_DESC);
-		if (check_mux(&ops[i], enq)) {
-			ret = enqueue_ldpc_enc_n_op_cb(q, &ops[i],
-					desc_idx, enq);
+		enq = check_mux(&ops[i], enq);
+		if (enq > 1) {
+			ret = enqueue_ldpc_enc_n_op_cb(q, &ops[i], desc_idx, enq);
 			if (ret < 0) {
 				acc_enqueue_invalid(q_data);
 				break;
-- 
2.37.1

[PATCH v3 4/7] baseband/acc: acc100 fix queue mapping to 64 bits

[Hernan Vargas]

Fix potential overflow for Q mapping extension to 64 bits.

Fixes: 32e8b7ea35d ("baseband/acc100: refactor to segregate common code")

Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
---
 drivers/baseband/acc/rte_acc100_pmd.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
index eb18cc2215..3660d3b240 100644
--- a/drivers/baseband/acc/rte_acc100_pmd.c
+++ b/drivers/baseband/acc/rte_acc100_pmd.c
@@ -662,7 +662,7 @@ acc100_find_free_queue_idx(struct rte_bbdev *dev,
 	for (aq_idx = 0; aq_idx < qtop->num_aqs_per_groups; aq_idx++) {
 		if (((d->q_assigned_bit_map[group_idx] >> aq_idx) & 0x1) == 0) {
 			/* Mark the Queue as assigned */
-			d->q_assigned_bit_map[group_idx] |= (1 << aq_idx);
+			d->q_assigned_bit_map[group_idx] |= (1ULL << aq_idx);
 			/* Report the AQ Index */
 			return (group_idx << ACC100_GRP_ID_SHIFT) + aq_idx;
 		}
-- 
2.37.1

[PATCH v3 5/7] baseband/acc: acc100 use define constant

[Hernan Vargas]

Use define constant ACC_HARQ_ALIGN_64B instead of hardcoded number.
No functional impact.

Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
---
 drivers/baseband/acc/rte_acc100_pmd.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
index 3660d3b240..ef1c488a1c 100644
--- a/drivers/baseband/acc/rte_acc100_pmd.c
+++ b/drivers/baseband/acc/rte_acc100_pmd.c
@@ -1390,7 +1390,7 @@ acc101_fcw_ld_fill(struct rte_bbdev_dec_op *op, struct acc_fcw_ld *fcw,
 		harq_in_length = RTE_MIN(harq_in_length, op->ldpc_dec.n_cb
 				- op->ldpc_dec.n_filler);
 		/* Alignment on next 64B - Already enforced from HC output */
-		harq_in_length = RTE_ALIGN_FLOOR(harq_in_length, 64);
+		harq_in_length = RTE_ALIGN_FLOOR(harq_in_length, ACC_HARQ_ALIGN_64B);
 		fcw->hcin_size0 = harq_in_length;
 		fcw->hcin_offset = 0;
 		fcw->hcin_size1 = 0;
@@ -1433,7 +1433,7 @@ acc101_fcw_ld_fill(struct rte_bbdev_dec_op *op, struct acc_fcw_ld *fcw,
 		/* Cannot exceed the pruned Ncb circular buffer */
 		harq_out_length = RTE_MIN(harq_out_length, ncb_p);
 		/* Alignment on next 64B */
-		harq_out_length = RTE_ALIGN_CEIL(harq_out_length, 64);
+		harq_out_length = RTE_ALIGN_CEIL(harq_out_length, ACC_HARQ_ALIGN_64B);
 		fcw->hcout_size0 = harq_out_length;
 		fcw->hcout_size1 = 0;
 		fcw->hcout_offset = 0;
-- 
2.37.1

[PATCH v3 6/7] baseband/acc: acc100 use desc helper functions

[Hernan Vargas]

Use the designated descriptor helper functions for descriptor address
calculations.
No functional impact.

Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
---
 drivers/baseband/acc/rte_acc100_pmd.c | 35 +++++++++------------------
 1 file changed, 12 insertions(+), 23 deletions(-)

diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
index ef1c488a1c..d3aaeb8f2c 100644
--- a/drivers/baseband/acc/rte_acc100_pmd.c
+++ b/drivers/baseband/acc/rte_acc100_pmd.c
@@ -2422,9 +2422,8 @@ enqueue_ldpc_enc_part_tb(struct acc_queue *q, struct rte_bbdev_enc_op *op,
 	struct rte_mbuf *output_head, *output;
 	int i, next_triplet;
 	struct rte_bbdev_op_ldpc_enc *enc = &op->ldpc_enc;
-	uint16_t desc_idx = ((q->sw_ring_head + total_enqueued_descs) & q->sw_ring_wrap_mask);
 
-	desc = q->ring_addr + desc_idx;
+	desc = acc_desc(q, total_enqueued_descs);
 	acc_fcw_le_fill(op, &desc->req.fcw_le, num_cbs, e);
 
 	/* This could be done at polling. */
@@ -2613,7 +2612,6 @@ enqueue_ldpc_enc_one_op_tb(struct acc_queue *q, struct rte_bbdev_enc_op *op,
 	}
 #endif
 	uint8_t num_a, num_b;
-	uint16_t desc_idx;
 	uint8_t r = op->ldpc_enc.tb_params.r;
 	uint8_t cab =  op->ldpc_enc.tb_params.cab;
 	union acc_dma_desc *desc;
@@ -2655,16 +2653,15 @@ enqueue_ldpc_enc_one_op_tb(struct acc_queue *q, struct rte_bbdev_enc_op *op,
 
 	return_descs = enq_descs - init_enq_descs;
 	/* Keep total number of CBs in first TB. */
-	desc_idx = ((q->sw_ring_head + init_enq_descs) & q->sw_ring_wrap_mask);
-	desc = q->ring_addr + desc_idx;
+	desc = acc_desc(q, init_enq_descs);
 	desc->req.cbs_in_tb = return_descs; /** Actual number of descriptors. */
 	desc->req.op_addr = op;
 
 	/* Set SDone on last CB descriptor for TB mode. */
-	desc_idx = ((q->sw_ring_head + enq_descs - 1) & q->sw_ring_wrap_mask);
-	desc = q->ring_addr + desc_idx;
+	desc = acc_desc(q, enq_descs - 1);
 	desc->req.sdone_enable = 1;
 	desc->req.op_addr = op;
+
 	return return_descs;
 }
 
@@ -3275,7 +3272,7 @@ enqueue_dec_one_op_tb(struct acc_queue *q, struct rte_bbdev_dec_op *op,
 		h_out_length, mbuf_total_left, seg_total_left;
 	struct rte_mbuf *input, *h_output_head, *h_output,
 		*s_output_head, *s_output;
-	uint16_t desc_idx, current_enqueued_cbs = 0;
+	uint16_t current_enqueued_cbs = 0;
 	uint64_t fcw_offset;
 
 #ifndef RTE_LIBRTE_BBDEV_SKIP_VALIDATE
@@ -3290,9 +3287,8 @@ enqueue_dec_one_op_tb(struct acc_queue *q, struct rte_bbdev_dec_op *op,
 	}
 #endif
 
-	desc_idx = acc_desc_idx(q, total_enqueued_cbs);
-	desc = q->ring_addr + desc_idx;
-	fcw_offset = (desc_idx << 8) + ACC_DESC_FCW_OFFSET;
+	desc = acc_desc(q, total_enqueued_cbs);
+	fcw_offset = (acc_desc_idx(q, total_enqueued_cbs) << 8) + ACC_DESC_FCW_OFFSET;
 	acc100_fcw_td_fill(op, &desc->req.fcw_td);
 
 	input = op->turbo_dec.input.data;
@@ -3777,7 +3773,6 @@ dequeue_enc_one_op_cb(struct acc_queue *q, struct rte_bbdev_enc_op **ref_op,
 
 	/* Clearing status, it will be set based on response */
 	op->status = 0;
-
 	op->status |= ((rsp.dma_err) ? (1 << RTE_BBDEV_DRV_ERROR) : 0);
 	op->status |= ((rsp.fcw_err) ? (1 << RTE_BBDEV_DRV_ERROR) : 0);
 
@@ -4023,8 +4018,8 @@ dequeue_dec_one_op_tb(struct acc_queue *q, struct rte_bbdev_dec_op **ref_op,
 		atom_desc.atom_hdr = __atomic_load_n((uint64_t *)desc,
 				__ATOMIC_RELAXED);
 		rsp.val = atom_desc.rsp.val;
-		rte_bbdev_log_debug("Resp. desc %p: %x", desc,
-				rsp.val);
+		rte_bbdev_log_debug("Resp. desc %p: %x r %d c %d\n",
+						desc, rsp.val, cb_idx, cbs_in_tb);
 
 		op->status |= ((rsp.input_err) ? (1 << RTE_BBDEV_DATA_ERROR) : 0);
 		op->status |= ((rsp.dma_err) ? (1 << RTE_BBDEV_DRV_ERROR) : 0);
@@ -4077,8 +4072,7 @@ acc100_dequeue_enc(struct rte_bbdev_queue_data *q_data,
 		return 0;
 	}
 #endif
-	op = (q->ring_addr + (q->sw_ring_tail &
-			q->sw_ring_wrap_mask))->req.op_addr;
+	op = acc_op_tail(q, 0);
 	if (unlikely(ops == NULL || op == NULL))
 		return 0;
 	cbm = op->turbo_enc.code_block_mode;
@@ -4120,7 +4114,6 @@ acc100_dequeue_ldpc_enc(struct rte_bbdev_queue_data *q_data,
 	uint16_t i, dequeued_ops = 0, dequeued_descs = 0;
 	int ret, cbm;
 	struct rte_bbdev_enc_op *op;
-	union acc_dma_desc *desc;
 
 	if (q == NULL)
 		return 0;
@@ -4128,10 +4121,7 @@ acc100_dequeue_ldpc_enc(struct rte_bbdev_queue_data *q_data,
 	if (unlikely(ops == 0))
 		return 0;
 #endif
-	desc = q->ring_addr + (q->sw_ring_tail & q->sw_ring_wrap_mask);
-	if (unlikely(desc == NULL))
-		return 0;
-	op = desc->req.op_addr;
+	op = acc_op_tail(q, 0);
 	if (unlikely(ops == NULL || op == NULL))
 		return 0;
 	cbm = op->ldpc_enc.code_block_mode;
@@ -4227,8 +4217,7 @@ acc100_dequeue_ldpc_dec(struct rte_bbdev_queue_data *q_data,
 	dequeue_num = RTE_MIN(avail, num);
 
 	for (i = 0; i < dequeue_num; ++i) {
-		op = (q->ring_addr + ((q->sw_ring_tail + dequeued_cbs)
-			& q->sw_ring_wrap_mask))->req.op_addr;
+		op = acc_op_tail(q, dequeued_cbs);
 		if (unlikely(op == NULL))
 			break;
 		if (op->ldpc_dec.code_block_mode == RTE_BBDEV_TRANSPORT_BLOCK)
-- 
2.37.1

[PATCH v3 7/7] baseband/acc: acc100 ignore missing mempools

[Hernan Vargas]

Mempool not strictly required to be present in op for processing.
Previous criteria too strict. No functional impact.

Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
---
 drivers/baseband/acc/rte_acc100_pmd.c | 9 ---------
 1 file changed, 9 deletions(-)

diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
index d3aaeb8f2c..0992cb5d1e 100644
--- a/drivers/baseband/acc/rte_acc100_pmd.c
+++ b/drivers/baseband/acc/rte_acc100_pmd.c
@@ -8,7 +8,6 @@
 #include <rte_log.h>
 #include <dev_driver.h>
 #include <rte_malloc.h>
-#include <rte_mempool.h>
 #include <rte_byteorder.h>
 #include <rte_errno.h>
 #include <rte_branch_prediction.h>
@@ -1830,10 +1829,6 @@ validate_enc_op(struct rte_bbdev_enc_op *op, struct acc_queue *q)
 	if (!validate_op_required(q))
 		return 0;
 
-	if (op->mempool == NULL) {
-		rte_bbdev_log(ERR, "Invalid mempool pointer");
-		return -1;
-	}
 	if (turbo_enc->input.data == NULL) {
 		rte_bbdev_log(ERR, "Invalid input pointer");
 		return -1;
@@ -2677,10 +2672,6 @@ validate_dec_op(struct rte_bbdev_dec_op *op, struct acc_queue *q)
 	if (!validate_op_required(q))
 		return 0;
 
-	if (op->mempool == NULL) {
-		rte_bbdev_log(ERR, "Invalid mempool pointer");
-		return -1;
-	}
 	if (turbo_dec->input.data == NULL) {
 		rte_bbdev_log(ERR, "Invalid input pointer");
 		return -1;
-- 
2.37.1

Re: [PATCH v3 0/7] baseband/acc: changes for 23.03

[Maxime Coquelin]

Hi Hernan,

On 1/12/23 20:36, Hernan Vargas wrote:
> v3: Remove unused mempool header.
> v2: Reordered commits and added one more fix.
> v1: Upstreaming ACC100 changes for 23.03.
> 
> Hernan Vargas (7):
>    baseband/acc: acc100 free harq layout pointer
>    baseband/acc: acc100 fix iteration counter in TB
>    baseband/acc: acc100 fix multiplexing multiple ops
>    baseband/acc: acc100 fix queue mapping to 64 bits
>    baseband/acc: acc100 use define constant
>    baseband/acc: acc100 use desc helper functions
>    baseband/acc: acc100 ignore missing mempools
> 
>   drivers/baseband/acc/rte_acc100_pmd.c | 65 +++++++++++----------------
>   1 file changed, 25 insertions(+), 40 deletions(-)
> 

As I mentioned in the first v3 sent, next time please include the R-by's
that were given on previous versions on patches that did changed.

For this time, I will handle it myself.

Regards,
Maxime

Re: [PATCH v3 1/7] baseband/acc: acc100 free harq layout pointer

[Maxime Coquelin]

On 1/12/23 20:36, Hernan Vargas wrote:
> Explicitly call rte_free for harq_layout pointer in dev_close function
> to prevent memory leak.
> 
> Fixes: ba2262fe16c ("baseband/acc100: fix close cleanup")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
> ---
>   drivers/baseband/acc/rte_acc100_pmd.c | 1 +
>   1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
> index b3164529dd..146a22bd64 100644
> --- a/drivers/baseband/acc/rte_acc100_pmd.c
> +++ b/drivers/baseband/acc/rte_acc100_pmd.c
> @@ -620,6 +620,7 @@ acc100_dev_close(struct rte_bbdev *dev)
>   		rte_free(d->tail_ptrs);
>   		rte_free(d->info_ring);
>   		rte_free(d->sw_rings_base);
> +		rte_free(d->harq_layout);
>   		d->sw_rings_base = NULL;
>   		d->tail_ptrs = NULL;
>   		d->info_ring = NULL;

Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime

Re: [PATCH v3 2/7] baseband/acc: acc100 fix iteration counter in TB

[Maxime Coquelin]

On 1/12/23 20:36, Hernan Vargas wrote:
> Use ldpc or turbo iteration counter based on the operation type.
> 
> Fixes: 5ad5060f8f7 ("baseband/acc100: add LDPC processing functions")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
> ---
>   drivers/baseband/acc/rte_acc100_pmd.c | 8 ++++++--
>   1 file changed, 6 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
> index 146a22bd64..a00314c957 100644
> --- a/drivers/baseband/acc/rte_acc100_pmd.c
> +++ b/drivers/baseband/acc/rte_acc100_pmd.c
> @@ -4033,8 +4033,12 @@ dequeue_dec_one_op_tb(struct acc_queue *q, struct rte_bbdev_dec_op **ref_op,
>   		/* CRC invalid if error exists */
>   		if (!op->status)
>   			op->status |= rsp.crc_status << RTE_BBDEV_CRC_ERROR;
> -		op->turbo_dec.iter_count = RTE_MAX((uint8_t) rsp.iter_cnt,
> -				op->turbo_dec.iter_count);
> +		if (q->op_type == RTE_BBDEV_OP_LDPC_DEC)
> +			op->ldpc_dec.iter_count = RTE_MAX((uint8_t) rsp.iter_cnt,
> +					op->ldpc_dec.iter_count);
> +		else
> +			op->turbo_dec.iter_count = RTE_MAX((uint8_t) rsp.iter_cnt,
> +					op->turbo_dec.iter_count);
>   
>   		/* Check if this is the last desc in batch (Atomic Queue) */
>   		if (desc->req.last_desc_in_batch) {

Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime

Re: [PATCH v3 3/7] baseband/acc: acc100 fix multiplexing multiple ops

[Maxime Coquelin]

On 1/12/23 20:36, Hernan Vargas wrote:
> Function to check if multiplexing is possible for multiple operations in
> one descriptor returns number of enqueued ops.
> 
> Fixes: 32e8b7ea35d ("baseband/acc100: refactor to segregate common code")
> 
> Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
> ---
>   drivers/baseband/acc/rte_acc100_pmd.c | 6 +++---
>   1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
> index a00314c957..eb18cc2215 100644
> --- a/drivers/baseband/acc/rte_acc100_pmd.c
> +++ b/drivers/baseband/acc/rte_acc100_pmd.c
> @@ -3421,9 +3421,9 @@ acc100_enqueue_ldpc_enc_cb(struct rte_bbdev_queue_data *q_data,
>   		}
>   		avail--;
>   		enq = RTE_MIN(left, ACC_MUX_5GDL_DESC);
> -		if (check_mux(&ops[i], enq)) {
> -			ret = enqueue_ldpc_enc_n_op_cb(q, &ops[i],
> -					desc_idx, enq);
> +		enq = check_mux(&ops[i], enq);
> +		if (enq > 1) {
> +			ret = enqueue_ldpc_enc_n_op_cb(q, &ops[i], desc_idx, enq);
>   			if (ret < 0) {
>   				acc_enqueue_invalid(q_data);
>   				break;

Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime

Re: [PATCH v3 4/7] baseband/acc: acc100 fix queue mapping to 64 bits

[Maxime Coquelin]

On 1/12/23 20:36, Hernan Vargas wrote:
> Fix potential overflow for Q mapping extension to 64 bits.
> 
> Fixes: 32e8b7ea35d ("baseband/acc100: refactor to segregate common code")
> 
> Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
> ---
>   drivers/baseband/acc/rte_acc100_pmd.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
> index eb18cc2215..3660d3b240 100644
> --- a/drivers/baseband/acc/rte_acc100_pmd.c
> +++ b/drivers/baseband/acc/rte_acc100_pmd.c
> @@ -662,7 +662,7 @@ acc100_find_free_queue_idx(struct rte_bbdev *dev,
>   	for (aq_idx = 0; aq_idx < qtop->num_aqs_per_groups; aq_idx++) {
>   		if (((d->q_assigned_bit_map[group_idx] >> aq_idx) & 0x1) == 0) {
>   			/* Mark the Queue as assigned */
> -			d->q_assigned_bit_map[group_idx] |= (1 << aq_idx);
> +			d->q_assigned_bit_map[group_idx] |= (1ULL << aq_idx);
>   			/* Report the AQ Index */
>   			return (group_idx << ACC100_GRP_ID_SHIFT) + aq_idx;
>   		}

Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime

Re: [PATCH v3 5/7] baseband/acc: acc100 use define constant

[Maxime Coquelin]

On 1/12/23 20:36, Hernan Vargas wrote:
> Use define constant ACC_HARQ_ALIGN_64B instead of hardcoded number.
> No functional impact.
> 
> Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
> ---
>   drivers/baseband/acc/rte_acc100_pmd.c | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
> index 3660d3b240..ef1c488a1c 100644
> --- a/drivers/baseband/acc/rte_acc100_pmd.c
> +++ b/drivers/baseband/acc/rte_acc100_pmd.c
> @@ -1390,7 +1390,7 @@ acc101_fcw_ld_fill(struct rte_bbdev_dec_op *op, struct acc_fcw_ld *fcw,
>   		harq_in_length = RTE_MIN(harq_in_length, op->ldpc_dec.n_cb
>   				- op->ldpc_dec.n_filler);
>   		/* Alignment on next 64B - Already enforced from HC output */
> -		harq_in_length = RTE_ALIGN_FLOOR(harq_in_length, 64);
> +		harq_in_length = RTE_ALIGN_FLOOR(harq_in_length, ACC_HARQ_ALIGN_64B);
>   		fcw->hcin_size0 = harq_in_length;
>   		fcw->hcin_offset = 0;
>   		fcw->hcin_size1 = 0;
> @@ -1433,7 +1433,7 @@ acc101_fcw_ld_fill(struct rte_bbdev_dec_op *op, struct acc_fcw_ld *fcw,
>   		/* Cannot exceed the pruned Ncb circular buffer */
>   		harq_out_length = RTE_MIN(harq_out_length, ncb_p);
>   		/* Alignment on next 64B */
> -		harq_out_length = RTE_ALIGN_CEIL(harq_out_length, 64);
> +		harq_out_length = RTE_ALIGN_CEIL(harq_out_length, ACC_HARQ_ALIGN_64B);
>   		fcw->hcout_size0 = harq_out_length;
>   		fcw->hcout_size1 = 0;
>   		fcw->hcout_offset = 0;

Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime

Re: [PATCH v3 6/7] baseband/acc: acc100 use desc helper functions

[Maxime Coquelin]

On 1/12/23 20:36, Hernan Vargas wrote:
> Use the designated descriptor helper functions for descriptor address
> calculations.
> No functional impact.
> 
> Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
> ---
>   drivers/baseband/acc/rte_acc100_pmd.c | 35 +++++++++------------------
>   1 file changed, 12 insertions(+), 23 deletions(-)
> 
> diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
> index ef1c488a1c..d3aaeb8f2c 100644
> --- a/drivers/baseband/acc/rte_acc100_pmd.c
> +++ b/drivers/baseband/acc/rte_acc100_pmd.c
> @@ -2422,9 +2422,8 @@ enqueue_ldpc_enc_part_tb(struct acc_queue *q, struct rte_bbdev_enc_op *op,
>   	struct rte_mbuf *output_head, *output;
>   	int i, next_triplet;
>   	struct rte_bbdev_op_ldpc_enc *enc = &op->ldpc_enc;
> -	uint16_t desc_idx = ((q->sw_ring_head + total_enqueued_descs) & q->sw_ring_wrap_mask);
>   
> -	desc = q->ring_addr + desc_idx;
> +	desc = acc_desc(q, total_enqueued_descs);
>   	acc_fcw_le_fill(op, &desc->req.fcw_le, num_cbs, e);
>   
>   	/* This could be done at polling. */
> @@ -2613,7 +2612,6 @@ enqueue_ldpc_enc_one_op_tb(struct acc_queue *q, struct rte_bbdev_enc_op *op,
>   	}
>   #endif
>   	uint8_t num_a, num_b;
> -	uint16_t desc_idx;
>   	uint8_t r = op->ldpc_enc.tb_params.r;
>   	uint8_t cab =  op->ldpc_enc.tb_params.cab;
>   	union acc_dma_desc *desc;
> @@ -2655,16 +2653,15 @@ enqueue_ldpc_enc_one_op_tb(struct acc_queue *q, struct rte_bbdev_enc_op *op,
>   
>   	return_descs = enq_descs - init_enq_descs;
>   	/* Keep total number of CBs in first TB. */
> -	desc_idx = ((q->sw_ring_head + init_enq_descs) & q->sw_ring_wrap_mask);
> -	desc = q->ring_addr + desc_idx;
> +	desc = acc_desc(q, init_enq_descs);
>   	desc->req.cbs_in_tb = return_descs; /** Actual number of descriptors. */
>   	desc->req.op_addr = op;
>   
>   	/* Set SDone on last CB descriptor for TB mode. */
> -	desc_idx = ((q->sw_ring_head + enq_descs - 1) & q->sw_ring_wrap_mask);
> -	desc = q->ring_addr + desc_idx;
> +	desc = acc_desc(q, enq_descs - 1);
>   	desc->req.sdone_enable = 1;
>   	desc->req.op_addr = op;
> +
>   	return return_descs;
>   }
>   
> @@ -3275,7 +3272,7 @@ enqueue_dec_one_op_tb(struct acc_queue *q, struct rte_bbdev_dec_op *op,
>   		h_out_length, mbuf_total_left, seg_total_left;
>   	struct rte_mbuf *input, *h_output_head, *h_output,
>   		*s_output_head, *s_output;
> -	uint16_t desc_idx, current_enqueued_cbs = 0;
> +	uint16_t current_enqueued_cbs = 0;
>   	uint64_t fcw_offset;
>   
>   #ifndef RTE_LIBRTE_BBDEV_SKIP_VALIDATE
> @@ -3290,9 +3287,8 @@ enqueue_dec_one_op_tb(struct acc_queue *q, struct rte_bbdev_dec_op *op,
>   	}
>   #endif
>   
> -	desc_idx = acc_desc_idx(q, total_enqueued_cbs);
> -	desc = q->ring_addr + desc_idx;
> -	fcw_offset = (desc_idx << 8) + ACC_DESC_FCW_OFFSET;
> +	desc = acc_desc(q, total_enqueued_cbs);
> +	fcw_offset = (acc_desc_idx(q, total_enqueued_cbs) << 8) + ACC_DESC_FCW_OFFSET;
>   	acc100_fcw_td_fill(op, &desc->req.fcw_td);
>   
>   	input = op->turbo_dec.input.data;
> @@ -3777,7 +3773,6 @@ dequeue_enc_one_op_cb(struct acc_queue *q, struct rte_bbdev_enc_op **ref_op,
>   
>   	/* Clearing status, it will be set based on response */
>   	op->status = 0;
> -
>   	op->status |= ((rsp.dma_err) ? (1 << RTE_BBDEV_DRV_ERROR) : 0);
>   	op->status |= ((rsp.fcw_err) ? (1 << RTE_BBDEV_DRV_ERROR) : 0);
>   
> @@ -4023,8 +4018,8 @@ dequeue_dec_one_op_tb(struct acc_queue *q, struct rte_bbdev_dec_op **ref_op,
>   		atom_desc.atom_hdr = __atomic_load_n((uint64_t *)desc,
>   				__ATOMIC_RELAXED);
>   		rsp.val = atom_desc.rsp.val;
> -		rte_bbdev_log_debug("Resp. desc %p: %x", desc,
> -				rsp.val);
> +		rte_bbdev_log_debug("Resp. desc %p: %x r %d c %d\n",
> +						desc, rsp.val, cb_idx, cbs_in_tb);
>   
>   		op->status |= ((rsp.input_err) ? (1 << RTE_BBDEV_DATA_ERROR) : 0);
>   		op->status |= ((rsp.dma_err) ? (1 << RTE_BBDEV_DRV_ERROR) : 0);
> @@ -4077,8 +4072,7 @@ acc100_dequeue_enc(struct rte_bbdev_queue_data *q_data,
>   		return 0;
>   	}
>   #endif
> -	op = (q->ring_addr + (q->sw_ring_tail &
> -			q->sw_ring_wrap_mask))->req.op_addr;
> +	op = acc_op_tail(q, 0);
>   	if (unlikely(ops == NULL || op == NULL))
>   		return 0;
>   	cbm = op->turbo_enc.code_block_mode;
> @@ -4120,7 +4114,6 @@ acc100_dequeue_ldpc_enc(struct rte_bbdev_queue_data *q_data,
>   	uint16_t i, dequeued_ops = 0, dequeued_descs = 0;
>   	int ret, cbm;
>   	struct rte_bbdev_enc_op *op;
> -	union acc_dma_desc *desc;
>   
>   	if (q == NULL)
>   		return 0;
> @@ -4128,10 +4121,7 @@ acc100_dequeue_ldpc_enc(struct rte_bbdev_queue_data *q_data,
>   	if (unlikely(ops == 0))
>   		return 0;
>   #endif
> -	desc = q->ring_addr + (q->sw_ring_tail & q->sw_ring_wrap_mask);
> -	if (unlikely(desc == NULL))
> -		return 0;
> -	op = desc->req.op_addr;
> +	op = acc_op_tail(q, 0);
>   	if (unlikely(ops == NULL || op == NULL))
>   		return 0;
>   	cbm = op->ldpc_enc.code_block_mode;
> @@ -4227,8 +4217,7 @@ acc100_dequeue_ldpc_dec(struct rte_bbdev_queue_data *q_data,
>   	dequeue_num = RTE_MIN(avail, num);
>   
>   	for (i = 0; i < dequeue_num; ++i) {
> -		op = (q->ring_addr + ((q->sw_ring_tail + dequeued_cbs)
> -			& q->sw_ring_wrap_mask))->req.op_addr;
> +		op = acc_op_tail(q, dequeued_cbs);
>   		if (unlikely(op == NULL))
>   			break;
>   		if (op->ldpc_dec.code_block_mode == RTE_BBDEV_TRANSPORT_BLOCK)

Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime

Re: [PATCH v3 7/7] baseband/acc: acc100 ignore missing mempools

[Maxime Coquelin]

On 1/12/23 20:36, Hernan Vargas wrote:
> Mempool not strictly required to be present in op for processing.
> Previous criteria too strict. No functional impact.
> 
> Signed-off-by: Hernan Vargas <hernan.vargas@intel.com>
> ---
>   drivers/baseband/acc/rte_acc100_pmd.c | 9 ---------
>   1 file changed, 9 deletions(-)
> 
> diff --git a/drivers/baseband/acc/rte_acc100_pmd.c b/drivers/baseband/acc/rte_acc100_pmd.c
> index d3aaeb8f2c..0992cb5d1e 100644
> --- a/drivers/baseband/acc/rte_acc100_pmd.c
> +++ b/drivers/baseband/acc/rte_acc100_pmd.c
> @@ -8,7 +8,6 @@
>   #include <rte_log.h>
>   #include <dev_driver.h>
>   #include <rte_malloc.h>
> -#include <rte_mempool.h>
>   #include <rte_byteorder.h>
>   #include <rte_errno.h>
>   #include <rte_branch_prediction.h>
> @@ -1830,10 +1829,6 @@ validate_enc_op(struct rte_bbdev_enc_op *op, struct acc_queue *q)
>   	if (!validate_op_required(q))
>   		return 0;
>   
> -	if (op->mempool == NULL) {
> -		rte_bbdev_log(ERR, "Invalid mempool pointer");
> -		return -1;
> -	}
>   	if (turbo_enc->input.data == NULL) {
>   		rte_bbdev_log(ERR, "Invalid input pointer");
>   		return -1;
> @@ -2677,10 +2672,6 @@ validate_dec_op(struct rte_bbdev_dec_op *op, struct acc_queue *q)
>   	if (!validate_op_required(q))
>   		return 0;
>   
> -	if (op->mempool == NULL) {
> -		rte_bbdev_log(ERR, "Invalid mempool pointer");
> -		return -1;
> -	}
>   	if (turbo_dec->input.data == NULL) {
>   		rte_bbdev_log(ERR, "Invalid input pointer");
>   		return -1;

Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime

Re: [PATCH v3 0/7] baseband/acc: changes for 23.03

[Maxime Coquelin]

On 1/12/23 20:36, Hernan Vargas wrote:
> v3: Remove unused mempool header.
> v2: Reordered commits and added one more fix.
> v1: Upstreaming ACC100 changes for 23.03.
> 
> Hernan Vargas (7):
>    baseband/acc: acc100 free harq layout pointer
>    baseband/acc: acc100 fix iteration counter in TB
>    baseband/acc: acc100 fix multiplexing multiple ops
>    baseband/acc: acc100 fix queue mapping to 64 bits
>    baseband/acc: acc100 use define constant
>    baseband/acc: acc100 use desc helper functions
>    baseband/acc: acc100 ignore missing mempools
> 
>   drivers/baseband/acc/rte_acc100_pmd.c | 65 +++++++++++----------------
>   1 file changed, 25 insertions(+), 40 deletions(-)
> 

Applied to dpdk-next-baseband/for-main.

Thanks,
Maxime