[v2] crypto/ipsec_mb: fix getting process ID per job
Checks
Commit Message
Currently, when using IPsec-mb 1.4+, the process ID is obtained for each
job in a burst with a call to getpid().
This system call uses too many CPU cycles, and is unnecessary per job.
Instead, set the process ID value per lcore.
This is read when processing the burst, instead of per job.
Fixes: 9593d83e5d88 ("crypto/ipsec_mb: fix aesni_mb multi-process session ID")
Cc: stable@dpdk.org
Signed-off-by: Ciara Power <ciara.power@intel.com>
---
drivers/crypto/ipsec_mb/pmd_aesni_mb.c | 22 ++++++++++++++++++----
1 file changed, 18 insertions(+), 4 deletions(-)
Comments
Acked-by: Kai Ji <kai.ji@intel.com>
> -----Original Message-----
> From: Power, Ciara <ciara.power@intel.com>
> Sent: Thursday, November 23, 2023 5:16 PM
> To: dev@dpdk.org
> Cc: thomas@monjalon.net; Ji, Kai <kai.ji@intel.com>; De Lara Guarch, Pablo
> <pablo.de.lara.guarch@intel.com>; Power, Ciara <ciara.power@intel.com>;
> stable@dpdk.org
> Subject: [PATCH v2] crypto/ipsec_mb: fix getting process ID per job
>
> Currently, when using IPsec-mb 1.4+, the process ID is obtained for each job in
> a burst with a call to getpid().
> This system call uses too many CPU cycles, and is unnecessary per job.
>
> Instead, set the process ID value per lcore.
> This is read when processing the burst, instead of per job.
>
> Fixes: 9593d83e5d88 ("crypto/ipsec_mb: fix aesni_mb multi-process session
> ID")
> Cc: stable@dpdk.org
>
> Signed-off-by: Ciara Power <ciara.power@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
> > Currently, when using IPsec-mb 1.4+, the process ID is obtained for each job in
> > a burst with a call to getpid().
> > This system call uses too many CPU cycles, and is unnecessary per job.
> >
> > Instead, set the process ID value per lcore.
> > This is read when processing the burst, instead of per job.
> >
> > Fixes: 9593d83e5d88 ("crypto/ipsec_mb: fix aesni_mb multi-process session
> > ID")
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Ciara Power <ciara.power@intel.com>
>
> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Applied, thanks.
@@ -6,6 +6,8 @@
#include "pmd_aesni_mb_priv.h"
+RTE_DEFINE_PER_LCORE(pid_t, pid);
+
struct aesni_mb_op_buf_data {
struct rte_mbuf *m;
uint32_t offset;
@@ -846,6 +848,7 @@ aesni_mb_session_configure(IMB_MGR *mb_mgr,
#if IMB_VERSION(1, 3, 0) < IMB_VERSION_NUM
sess->session_id = imb_set_session(mb_mgr, &sess->template_job);
sess->pid = getpid();
+ RTE_PER_LCORE(pid) = sess->pid;
#endif
return 0;
@@ -1503,7 +1506,7 @@ aesni_mb_digest_appended_in_src(struct rte_crypto_op *op, IMB_JOB *job,
static inline int
set_mb_job_params(IMB_JOB *job, struct ipsec_mb_qp *qp,
struct rte_crypto_op *op, uint8_t *digest_idx,
- IMB_MGR *mb_mgr)
+ IMB_MGR *mb_mgr, pid_t pid)
{
struct rte_mbuf *m_src = op->sym->m_src, *m_dst;
struct aesni_mb_qp_data *qp_data = ipsec_mb_get_qp_private_data(qp);
@@ -1517,6 +1520,10 @@ set_mb_job_params(IMB_JOB *job, struct ipsec_mb_qp *qp,
uint8_t sgl = 0;
uint8_t lb_sgl = 0;
+#if IMB_VERSION(1, 3, 0) >= IMB_VERSION_NUM
+ (void) pid;
+#endif
+
session = ipsec_mb_get_session_private(qp, op);
if (session == NULL) {
op->status = RTE_CRYPTO_OP_STATUS_INVALID_SESSION;
@@ -1527,7 +1534,7 @@ set_mb_job_params(IMB_JOB *job, struct ipsec_mb_qp *qp,
session->template_job.cipher_mode;
#if IMB_VERSION(1, 3, 0) < IMB_VERSION_NUM
- if (session->pid != getpid()) {
+ if (session->pid != pid) {
memcpy(job, &session->template_job, sizeof(IMB_JOB));
imb_set_session(mb_mgr, job);
} else if (job->session_id != session->session_id)
@@ -2136,6 +2143,7 @@ aesni_mb_dequeue_burst(void *queue_pair, struct rte_crypto_op **ops,
int retval, processed_jobs = 0;
uint16_t i, nb_jobs;
IMB_JOB *jobs[IMB_MAX_BURST_SIZE] = {NULL};
+ pid_t pid;
if (unlikely(nb_ops == 0 || mb_mgr == NULL))
return 0;
@@ -2176,6 +2184,11 @@ aesni_mb_dequeue_burst(void *queue_pair, struct rte_crypto_op **ops,
continue;
}
+ if (!RTE_PER_LCORE(pid))
+ RTE_PER_LCORE(pid) = getpid();
+
+ pid = RTE_PER_LCORE(pid);
+
/*
* Get the next operations to process from ingress queue.
* There is no need to return the job to the IMB_MGR
@@ -2194,7 +2207,7 @@ aesni_mb_dequeue_burst(void *queue_pair, struct rte_crypto_op **ops,
&digest_idx);
else
retval = set_mb_job_params(job, qp, op,
- &digest_idx, mb_mgr);
+ &digest_idx, mb_mgr, pid);
if (unlikely(retval != 0)) {
qp->stats.dequeue_err_count++;
@@ -2317,6 +2330,7 @@ aesni_mb_dequeue_burst(void *queue_pair, struct rte_crypto_op **ops,
struct rte_crypto_op *op;
IMB_JOB *job;
int retval, processed_jobs = 0;
+ pid_t pid = 0;
if (unlikely(nb_ops == 0 || mb_mgr == NULL))
return 0;
@@ -2353,7 +2367,7 @@ aesni_mb_dequeue_burst(void *queue_pair, struct rte_crypto_op **ops,
&digest_idx);
else
retval = set_mb_job_params(job, qp, op,
- &digest_idx, mb_mgr);
+ &digest_idx, mb_mgr, pid);
if (unlikely(retval != 0)) {
qp->stats.dequeue_err_count++;