Re: [dpdk-dev] [PATCH 5/6] crypto/dpaa_sec: fix session qp attach/detach

[Kevin Traynor <ktraynor@redhat.com>]

On 25/04/2019 17:24, Kevin Traynor wrote:
> On 27/03/2019 11:53, Akhil Goyal wrote:
>> session inq and qp are assigned for each core from which the
>> packets arrive. This was not correctly handled while supporting
>> multiple sessions per queue pair.
>> This patch fixes the attach and detach of queues for each core.
>>
>> Fixes: e79416d10fa3 ("crypto/dpaa_sec: support multiple sessions per queue pair")
>> Cc: stable@dpdk.org
> 
> Hi, this will not backport to 18.11 branch as:
> 4e694fe51171dcdbe94019189a0240833b45c943
> Author: Akhil Goyal <akhil.goyal@nxp.com>
> Date:   Wed Jan 9 15:14:17 2019 +0000
> 
>     crypto/dpaa_sec: support same session flows on multi-cores
> 
> was added after 18.11 and dpaa_sec_sym_session_clear() is quite
> different. Please send a backport of the bugfix for 18.11 branch, or let
> me know if it's not needed.
> 
> thanks,
> Kevin.
> 
>> Signed-off-by: Akhil Goyal <akhil.goyal@nxp.com>
>> ---
>>  drivers/crypto/dpaa_sec/dpaa_sec.c | 17 ++++++++++-------
>>  1 file changed, 10 insertions(+), 7 deletions(-)
>>
>> diff --git a/drivers/crypto/dpaa_sec/dpaa_sec.c b/drivers/crypto/dpaa_sec/dpaa_sec.c
>> index cb99be4e1..8305f19a3 100644
>> --- a/drivers/crypto/dpaa_sec/dpaa_sec.c
>> +++ b/drivers/crypto/dpaa_sec/dpaa_sec.c
>> @@ -1,7 +1,7 @@
>>  /* SPDX-License-Identifier: BSD-3-Clause
>>   *
>>   *   Copyright (c) 2016 Freescale Semiconductor, Inc. All rights reserved.
>> - *   Copyright 2017-2018 NXP
>> + *   Copyright 2017-2019 NXP
>>   *
>>   */
>>  
>> @@ -1940,13 +1940,13 @@ dpaa_sec_attach_rxq(struct dpaa_sec_dev_private *qi)
>>  {
>>  	unsigned int i;
>>  
>> -	for (i = 0; i < qi->max_nb_sessions; i++) {
>> +	for (i = 0; i < qi->max_nb_sessions * MAX_DPAA_CORES; i++) {
>>  		if (qi->inq_attach[i] == 0) {

Btw, without knowing the crypto code, the use of max_nb_sessions in for
loops with fixed size arrays being indexed looks suspect to me. In the
snippet above

unsigned char inq_attach[RTE_DPAA_MAX_RX_QUEUE];
which means inq_attach[4096]

At least in one place I can see
.max_nb_sessions = MRVL_PMD_DEFAULT_MAX_NB_SESSIONS
which makes max_nb_sessions=2048, while MAX_DPAA_CORES=4

>>  			qi->inq_attach[i] = 1;
>>  			return &qi->inq[i];
>>  		}
>>  	}
>> -	DPAA_SEC_WARN("All ses session in use %x", qi->max_nb_sessions);
>> +	DPAA_SEC_WARN("All session in use %u", qi->max_nb_sessions);
>>  
>>  	return NULL;
>>  }
>> @@ -2115,7 +2115,7 @@ dpaa_sec_sym_session_clear(struct rte_cryptodev *dev,
>>  		struct rte_cryptodev_sym_session *sess)
>>  {
>>  	struct dpaa_sec_dev_private *qi = dev->data->dev_private;
>> -	uint8_t index = dev->driver_id;
>> +	uint8_t index = dev->driver_id, i;
>>  	void *sess_priv = get_sym_session_private_data(sess, index);
>>  
>>  	PMD_INIT_FUNC_TRACE();
>> @@ -2125,9 +2125,12 @@ dpaa_sec_sym_session_clear(struct rte_cryptodev *dev,
>>  	if (sess_priv) {
>>  		struct rte_mempool *sess_mp = rte_mempool_from_obj(sess_priv);
>>  
>> -		if (s->inq[rte_lcore_id() % MAX_DPAA_CORES])
>> -			dpaa_sec_detach_rxq(qi,
>> -				s->inq[rte_lcore_id() % MAX_DPAA_CORES]);
>> +		for (i = 0; i < MAX_DPAA_CORES; i++) {
>> +			if (s->inq[i])
>> +				dpaa_sec_detach_rxq(qi, s->inq[i]);
>> +			s->inq[i] = NULL;
>> +			s->qp[i] = NULL;
>> +		}
>>  		rte_free(s->cipher_key.data);
>>  		rte_free(s->auth_key.data);
>>  		memset(s, 0, sizeof(dpaa_sec_session));
>>
>