* [dpdk-dev] [PATCH v1] net/ice: fix memzone leak when device init failed
@ 2021-08-13 6:21 Haiyue Wang
2021-08-17 9:19 ` David Marchand
2021-08-29 11:01 ` Zhang, Qi Z
0 siblings, 2 replies; 5+ messages in thread
From: Haiyue Wang @ 2021-08-13 6:21 UTC (permalink / raw)
To: dev
Cc: qi.z.zhang, Haiyue Wang, stable, David Marchand, Qiming Yang,
Xiaolong Ye, Beilei Xing, Ying A Wang
When flow engine initialization or FXP resource reset failed, it needs
to free the memory zone and unregister the interrupt callback.
Bugzilla ID: 752
Fixes: 84dc7a95a2d3 ("net/ice: enable flow director engine")
Fixes: 7615a6895009 ("net/ice: rework for generic flow enabling")
Fixes: 7edc7158d771 ("net/ice: cleanup RSS/FDIR profile on device init")
Cc: stable@dpdk.org
Reported-by: David Marchand <david.marchand@redhat.com>
Signed-off-by: Haiyue Wang <haiyue.wang@intel.com>
---
drivers/net/ice/ice_ethdev.c | 10 ++++++++--
drivers/net/ice/ice_fdir_filter.c | 2 ++
2 files changed, 10 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index 64ee569525..8d62b84805 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -2139,20 +2139,26 @@ ice_dev_init(struct rte_eth_dev *dev)
ret = ice_flow_init(ad);
if (ret) {
PMD_INIT_LOG(ERR, "Failed to initialize flow");
- return ret;
+ goto err_flow_init;
}
}
ret = ice_reset_fxp_resource(hw);
if (ret) {
PMD_INIT_LOG(ERR, "Failed to reset fxp resource");
- return ret;
+ goto err_flow_init;
}
pf->supported_rxdid = ice_get_supported_rxdid(hw);
return 0;
+err_flow_init:
+ ice_flow_uninit(ad);
+ rte_intr_disable(intr_handle);
+ ice_pf_disable_irq0(hw);
+ rte_intr_callback_unregister(intr_handle,
+ ice_interrupt_handler, dev);
err_pf_setup:
ice_res_pool_destroy(&pf->msix_pool);
err_msix_pool_init:
diff --git a/drivers/net/ice/ice_fdir_filter.c b/drivers/net/ice/ice_fdir_filter.c
index 82adb1fc8b..7ba65b9b04 100644
--- a/drivers/net/ice/ice_fdir_filter.c
+++ b/drivers/net/ice/ice_fdir_filter.c
@@ -651,8 +651,10 @@ ice_fdir_teardown(struct ice_pf *pf)
ice_tx_queue_release(pf->fdir.txq);
pf->fdir.txq = NULL;
+ rte_eth_dma_zone_free(eth_dev, "fdir_tx_ring", ICE_FDIR_QUEUE_ID);
ice_rx_queue_release(pf->fdir.rxq);
pf->fdir.rxq = NULL;
+ rte_eth_dma_zone_free(eth_dev, "fdir_rx_ring", ICE_FDIR_QUEUE_ID);
ice_fdir_prof_rm_all(pf);
ice_fdir_prof_free(hw);
ice_release_vsi(vsi);
--
2.32.0
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [dpdk-dev] [PATCH v1] net/ice: fix memzone leak when device init failed
2021-08-13 6:21 [dpdk-dev] [PATCH v1] net/ice: fix memzone leak when device init failed Haiyue Wang
@ 2021-08-17 9:19 ` David Marchand
2021-08-18 0:46 ` Wang, Haiyue
2021-08-29 11:01 ` Zhang, Qi Z
1 sibling, 1 reply; 5+ messages in thread
From: David Marchand @ 2021-08-17 9:19 UTC (permalink / raw)
To: Haiyue Wang
Cc: dev, Qi Zhang, dpdk stable, Qiming Yang, Xiaolong Ye,
Beilei Xing, Ying A Wang
On Fri, Aug 13, 2021 at 8:45 AM Haiyue Wang <haiyue.wang@intel.com> wrote:
>
> When flow engine initialization or FXP resource reset failed, it needs
> to free the memory zone and unregister the interrupt callback.
>
> Bugzilla ID: 752
> Fixes: 84dc7a95a2d3 ("net/ice: enable flow director engine")
> Fixes: 7615a6895009 ("net/ice: rework for generic flow enabling")
> Fixes: 7edc7158d771 ("net/ice: cleanup RSS/FDIR profile on device init")
> Cc: stable@dpdk.org
>
> Reported-by: David Marchand <david.marchand@redhat.com>
> Signed-off-by: Haiyue Wang <haiyue.wang@intel.com>
> ---
> drivers/net/ice/ice_ethdev.c | 10 ++++++++--
> drivers/net/ice/ice_fdir_filter.c | 2 ++
> 2 files changed, 10 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
> index 64ee569525..8d62b84805 100644
> --- a/drivers/net/ice/ice_ethdev.c
> +++ b/drivers/net/ice/ice_ethdev.c
> @@ -2139,20 +2139,26 @@ ice_dev_init(struct rte_eth_dev *dev)
> ret = ice_flow_init(ad);
> if (ret) {
> PMD_INIT_LOG(ERR, "Failed to initialize flow");
> - return ret;
> + goto err_flow_init;
Is it safe to call flow engine uninit callbacks when ice_flow_init() fails?
> }
> }
>
> ret = ice_reset_fxp_resource(hw);
> if (ret) {
> PMD_INIT_LOG(ERR, "Failed to reset fxp resource");
> - return ret;
> + goto err_flow_init;
> }
>
> pf->supported_rxdid = ice_get_supported_rxdid(hw);
>
> return 0;
>
> +err_flow_init:
> + ice_flow_uninit(ad);
> + rte_intr_disable(intr_handle);
> + ice_pf_disable_irq0(hw);
> + rte_intr_callback_unregister(intr_handle,
> + ice_interrupt_handler, dev);
> err_pf_setup:
> ice_res_pool_destroy(&pf->msix_pool);
> err_msix_pool_init:
--
David Marchand
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [dpdk-dev] [PATCH v1] net/ice: fix memzone leak when device init failed
2021-08-17 9:19 ` David Marchand
@ 2021-08-18 0:46 ` Wang, Haiyue
2021-08-18 7:07 ` David Marchand
0 siblings, 1 reply; 5+ messages in thread
From: Wang, Haiyue @ 2021-08-18 0:46 UTC (permalink / raw)
To: David Marchand
Cc: dev, Zhang, Qi Z, dpdk stable, Yang, Qiming, Xiaolong Ye, Xing,
Beilei, Wang, Ying A
> -----Original Message-----
> From: David Marchand <david.marchand@redhat.com>
> Sent: Tuesday, August 17, 2021 17:19
> To: Wang, Haiyue <haiyue.wang@intel.com>
> Cc: dev <dev@dpdk.org>; Zhang, Qi Z <qi.z.zhang@intel.com>; dpdk stable <stable@dpdk.org>; Yang,
> Qiming <qiming.yang@intel.com>; Xiaolong Ye <xiaolong.ye@intel.com>; Xing, Beilei
> <beilei.xing@intel.com>; Wang, Ying A <ying.a.wang@intel.com>
> Subject: Re: [PATCH v1] net/ice: fix memzone leak when device init failed
>
> On Fri, Aug 13, 2021 at 8:45 AM Haiyue Wang <haiyue.wang@intel.com> wrote:
> >
> > When flow engine initialization or FXP resource reset failed, it needs
> > to free the memory zone and unregister the interrupt callback.
> >
> > Bugzilla ID: 752
> > Fixes: 84dc7a95a2d3 ("net/ice: enable flow director engine")
> > Fixes: 7615a6895009 ("net/ice: rework for generic flow enabling")
> > Fixes: 7edc7158d771 ("net/ice: cleanup RSS/FDIR profile on device init")
> > Cc: stable@dpdk.org
> >
> > Reported-by: David Marchand <david.marchand@redhat.com>
> > Signed-off-by: Haiyue Wang <haiyue.wang@intel.com>
> > ---
> > drivers/net/ice/ice_ethdev.c | 10 ++++++++--
> > drivers/net/ice/ice_fdir_filter.c | 2 ++
> > 2 files changed, 10 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
> > index 64ee569525..8d62b84805 100644
> > --- a/drivers/net/ice/ice_ethdev.c
> > +++ b/drivers/net/ice/ice_ethdev.c
> > @@ -2139,20 +2139,26 @@ ice_dev_init(struct rte_eth_dev *dev)
> > ret = ice_flow_init(ad);
> > if (ret) {
> > PMD_INIT_LOG(ERR, "Failed to initialize flow");
> > - return ret;
> > + goto err_flow_init;
>
> Is it safe to call flow engine uninit callbacks when ice_flow_init() fails?
If each engine->init/uninit handles its internal setting correctly, yes,
it's safe, if not, this single engine has BUG, let's fix it. ;-)
int
ice_flow_init(struct ice_adapter *ad)
{
TAILQ_FOREACH_SAFE(engine, &engine_list, node, temp) {
if (engine->init == NULL) {
PMD_INIT_LOG(ERR, "Invalid engine type (%d)",
engine->type);
return -ENOTSUP;
}
ret = engine->init(ad);
if (ret) {
PMD_INIT_LOG(ERR, "Failed to initialize engine %d",
engine->type);
return ret;
}
}
}
void
ice_flow_uninit(struct ice_adapter *ad)
{
struct ice_pf *pf = &ad->pf;
struct ice_flow_engine *engine;
struct rte_flow *p_flow;
struct ice_flow_parser_node *p_parser;
void *temp;
TAILQ_FOREACH_SAFE(engine, &engine_list, node, temp) {
if (engine->uninit)
engine->uninit(ad);
}
}
>
>
> > }
> > }
> >
> > ret = ice_reset_fxp_resource(hw);
> > if (ret) {
> > PMD_INIT_LOG(ERR, "Failed to reset fxp resource");
> > - return ret;
> > + goto err_flow_init;
> > }
> >
> > pf->supported_rxdid = ice_get_supported_rxdid(hw);
> >
> > return 0;
> >
> > +err_flow_init:
> > + ice_flow_uninit(ad);
> > + rte_intr_disable(intr_handle);
> > + ice_pf_disable_irq0(hw);
> > + rte_intr_callback_unregister(intr_handle,
> > + ice_interrupt_handler, dev);
> > err_pf_setup:
> > ice_res_pool_destroy(&pf->msix_pool);
> > err_msix_pool_init:
>
>
> --
> David Marchand
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [dpdk-dev] [PATCH v1] net/ice: fix memzone leak when device init failed
2021-08-18 0:46 ` Wang, Haiyue
@ 2021-08-18 7:07 ` David Marchand
0 siblings, 0 replies; 5+ messages in thread
From: David Marchand @ 2021-08-18 7:07 UTC (permalink / raw)
To: Wang, Haiyue
Cc: dev, Zhang, Qi Z, dpdk stable, Yang, Qiming, Xiaolong Ye, Xing,
Beilei, Wang, Ying A
On Wed, Aug 18, 2021 at 2:46 AM Wang, Haiyue <haiyue.wang@intel.com> wrote:
>
> > -----Original Message-----
> > From: David Marchand <david.marchand@redhat.com>
> > Sent: Tuesday, August 17, 2021 17:19
> > To: Wang, Haiyue <haiyue.wang@intel.com>
> > Cc: dev <dev@dpdk.org>; Zhang, Qi Z <qi.z.zhang@intel.com>; dpdk stable <stable@dpdk.org>; Yang,
> > Qiming <qiming.yang@intel.com>; Xiaolong Ye <xiaolong.ye@intel.com>; Xing, Beilei
> > <beilei.xing@intel.com>; Wang, Ying A <ying.a.wang@intel.com>
> > Subject: Re: [PATCH v1] net/ice: fix memzone leak when device init failed
> >
> > On Fri, Aug 13, 2021 at 8:45 AM Haiyue Wang <haiyue.wang@intel.com> wrote:
> > >
> > > When flow engine initialization or FXP resource reset failed, it needs
> > > to free the memory zone and unregister the interrupt callback.
> > >
> > > Bugzilla ID: 752
> > > Fixes: 84dc7a95a2d3 ("net/ice: enable flow director engine")
> > > Fixes: 7615a6895009 ("net/ice: rework for generic flow enabling")
> > > Fixes: 7edc7158d771 ("net/ice: cleanup RSS/FDIR profile on device init")
> > > Cc: stable@dpdk.org
> > >
> > > Reported-by: David Marchand <david.marchand@redhat.com>
> > > Signed-off-by: Haiyue Wang <haiyue.wang@intel.com>
> > > ---
> > > drivers/net/ice/ice_ethdev.c | 10 ++++++++--
> > > drivers/net/ice/ice_fdir_filter.c | 2 ++
> > > 2 files changed, 10 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
> > > index 64ee569525..8d62b84805 100644
> > > --- a/drivers/net/ice/ice_ethdev.c
> > > +++ b/drivers/net/ice/ice_ethdev.c
> > > @@ -2139,20 +2139,26 @@ ice_dev_init(struct rte_eth_dev *dev)
> > > ret = ice_flow_init(ad);
> > > if (ret) {
> > > PMD_INIT_LOG(ERR, "Failed to initialize flow");
> > > - return ret;
> > > + goto err_flow_init;
> >
> > Is it safe to call flow engine uninit callbacks when ice_flow_init() fails?
>
> If each engine->init/uninit handles its internal setting correctly, yes,
> it's safe, if not, this single engine has BUG, let's fix it. ;-)
That was my understanding, but I preferred to ask.
Then, patch lgtm, thanks Haiyue.
--
David Marchand
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [dpdk-dev] [PATCH v1] net/ice: fix memzone leak when device init failed
2021-08-13 6:21 [dpdk-dev] [PATCH v1] net/ice: fix memzone leak when device init failed Haiyue Wang
2021-08-17 9:19 ` David Marchand
@ 2021-08-29 11:01 ` Zhang, Qi Z
1 sibling, 0 replies; 5+ messages in thread
From: Zhang, Qi Z @ 2021-08-29 11:01 UTC (permalink / raw)
To: Wang, Haiyue, dev
Cc: stable, David Marchand, Yang, Qiming, Xiaolong Ye, Xing, Beilei,
Wang, Ying A
> -----Original Message-----
> From: Wang, Haiyue <haiyue.wang@intel.com>
> Sent: Friday, August 13, 2021 2:22 PM
> To: dev@dpdk.org
> Cc: Zhang, Qi Z <qi.z.zhang@intel.com>; Wang, Haiyue
> <haiyue.wang@intel.com>; stable@dpdk.org; David Marchand
> <david.marchand@redhat.com>; Yang, Qiming <qiming.yang@intel.com>;
> Xiaolong Ye <xiaolong.ye@intel.com>; Xing, Beilei <beilei.xing@intel.com>;
> Wang, Ying A <ying.a.wang@intel.com>
> Subject: [PATCH v1] net/ice: fix memzone leak when device init failed
>
> When flow engine initialization or FXP resource reset failed, it needs to free the
> memory zone and unregister the interrupt callback.
>
> Bugzilla ID: 752
> Fixes: 84dc7a95a2d3 ("net/ice: enable flow director engine")
> Fixes: 7615a6895009 ("net/ice: rework for generic flow enabling")
> Fixes: 7edc7158d771 ("net/ice: cleanup RSS/FDIR profile on device init")
> Cc: stable@dpdk.org
>
> Reported-by: David Marchand <david.marchand@redhat.com>
> Signed-off-by: Haiyue Wang <haiyue.wang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Applied dpdk-next-net-intel.
Thanks
Qi
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2021-08-29 11:01 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-08-13 6:21 [dpdk-dev] [PATCH v1] net/ice: fix memzone leak when device init failed Haiyue Wang
2021-08-17 9:19 ` David Marchand
2021-08-18 0:46 ` Wang, Haiyue
2021-08-18 7:07 ` David Marchand
2021-08-29 11:01 ` Zhang, Qi Z
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).