* [dpdk-dev] [PATCH v2] net/ice: code clean for queue interrupt config
2020-03-15 3:22 [dpdk-dev] [PATCH] net/ice: fix wrong interrupt vector overwritten Qi Zhang
@ 2020-03-15 7:40 ` Qi Zhang
2020-03-15 9:01 ` [dpdk-dev] [PATCH v3] " Qi Zhang
1 sibling, 0 replies; 3+ messages in thread
From: Qi Zhang @ 2020-03-15 7:40 UTC (permalink / raw)
To: beilei.xing, yahui.cao; +Cc: xiaolong.ye, dev, Qi Zhang
The patch decouple the intr_handle from ice_vsi_queue_bind_intr which
is shared by data Rx queue and fdir Rx queue's interrupt binding.
Though there is no issue in current implemenation of
ice_vsi_queue_bind_intr, but this rely on the assumption that the fdir
setup (when bind interrrupt for fdir queue) must be called before
dev_start (when bind interrupt for data queue), which is not a good
practise. More detail: if we deferred fdir setup to after dev_start,
it is possible when configure a fdir queue, a recorded vector for Rx
queue 0 be overwritten by the fdir queue's vector number, this may cause
interrupt Rx mode does not work on the Rx queue 0, since the vector
number be used by rte_eth_dev_rx_intr_ctl is corrupted.
The patch also apply the same code decoupling on
ice_vsi_enable|disable_queue_intr to make all queue interrupt API more
consistent.
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
---
v2:
- rewrite the commit, actually this is not a fix, but code clean.
drivers/net/ice/ice_ethdev.c | 52 ++++++++++++++++-----------------------
drivers/net/ice/ice_ethdev.h | 9 ++++---
drivers/net/ice/ice_fdir_filter.c | 6 ++---
3 files changed, 30 insertions(+), 37 deletions(-)
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index 85ef83e92..48ce258ca 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -2318,11 +2318,8 @@ ice_release_vsi(struct ice_vsi *vsi)
}
void
-ice_vsi_disable_queues_intr(struct ice_vsi *vsi)
+ice_vsi_disable_queues_intr(struct ice_vsi *vsi, bool use_misc)
{
- struct rte_eth_dev *dev = vsi->adapter->eth_dev;
- struct rte_pci_device *pci_dev = ICE_DEV_TO_PCI(dev);
- struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
struct ice_hw *hw = ICE_VSI_TO_HW(vsi);
uint16_t msix_intr, i;
@@ -2333,7 +2330,7 @@ ice_vsi_disable_queues_intr(struct ice_vsi *vsi)
rte_wmb();
}
- if (rte_intr_allow_others(intr_handle))
+ if (use_misc)
/* vfio-pci */
for (i = 0; i < vsi->nb_msix; i++) {
msix_intr = vsi->msix_intr + i;
@@ -2368,7 +2365,8 @@ ice_dev_stop(struct rte_eth_dev *dev)
ice_tx_queue_stop(dev, i);
/* disable all queue interrupts */
- ice_vsi_disable_queues_intr(main_vsi);
+ ice_vsi_disable_queues_intr(main_vsi,
+ !rte_intr_allow_others(intr_handle));
/* Clear all queues and release mbufs */
ice_clear_queues(dev);
@@ -2619,16 +2617,15 @@ __vsi_queues_bind_intr(struct ice_vsi *vsi, uint16_t msix_vect,
}
void
-ice_vsi_queues_bind_intr(struct ice_vsi *vsi)
+ice_vsi_queues_bind_intr(struct ice_vsi *vsi,
+ bool use_misc,
+ int intr_num_max,
+ int *intr_vec)
{
- struct rte_eth_dev *dev = vsi->adapter->eth_dev;
- struct rte_pci_device *pci_dev = ICE_DEV_TO_PCI(dev);
- struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
struct ice_hw *hw = ICE_VSI_TO_HW(vsi);
uint16_t msix_vect = vsi->msix_intr;
- uint16_t nb_msix = RTE_MIN(vsi->nb_msix, intr_handle->nb_efd);
+ uint16_t nb_msix = RTE_MIN(vsi->nb_msix, intr_num_max);
uint16_t queue_idx = 0;
- int record = 0;
int i;
/* clear Rx/Tx queue interrupt */
@@ -2637,15 +2634,9 @@ ice_vsi_queues_bind_intr(struct ice_vsi *vsi)
ICE_WRITE_REG(hw, QINT_RQCTL(vsi->base_queue + i), 0);
}
- /* PF bind interrupt */
- if (rte_intr_dp_is_en(intr_handle)) {
- queue_idx = 0;
- record = 1;
- }
-
for (i = 0; i < vsi->nb_used_qps; i++) {
if (nb_msix <= 1) {
- if (!rte_intr_allow_others(intr_handle))
+ if (use_misc)
msix_vect = ICE_MISC_VEC_ID;
/* uio mapping all queue to one msix_vect */
@@ -2653,9 +2644,8 @@ ice_vsi_queues_bind_intr(struct ice_vsi *vsi)
vsi->base_queue + i,
vsi->nb_used_qps - i);
- for (; !!record && i < vsi->nb_used_qps; i++)
- intr_handle->intr_vec[queue_idx + i] =
- msix_vect;
+ for (; intr_vec && i < vsi->nb_used_qps; i++)
+ intr_vec[queue_idx + i] = msix_vect;
break;
}
@@ -2663,8 +2653,8 @@ ice_vsi_queues_bind_intr(struct ice_vsi *vsi)
__vsi_queues_bind_intr(vsi, msix_vect,
vsi->base_queue + i, 1);
- if (!!record)
- intr_handle->intr_vec[queue_idx + i] = msix_vect;
+ if (intr_vec)
+ intr_vec[queue_idx + i] = msix_vect;
msix_vect++;
nb_msix--;
@@ -2672,15 +2662,12 @@ ice_vsi_queues_bind_intr(struct ice_vsi *vsi)
}
void
-ice_vsi_enable_queues_intr(struct ice_vsi *vsi)
+ice_vsi_enable_queues_intr(struct ice_vsi *vsi, bool use_misc)
{
- struct rte_eth_dev *dev = vsi->adapter->eth_dev;
- struct rte_pci_device *pci_dev = ICE_DEV_TO_PCI(dev);
- struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
struct ice_hw *hw = ICE_VSI_TO_HW(vsi);
uint16_t msix_intr, i;
- if (rte_intr_allow_others(intr_handle))
+ if (use_misc)
for (i = 0; i < vsi->nb_used_qps; i++) {
msix_intr = vsi->msix_intr + i;
ICE_WRITE_REG(hw, GLINT_DYN_CTL(msix_intr),
@@ -2736,10 +2723,13 @@ ice_rxq_intr_setup(struct rte_eth_dev *dev)
/* Map queues with MSIX interrupt */
vsi->nb_used_qps = dev->data->nb_rx_queues;
- ice_vsi_queues_bind_intr(vsi);
+ ice_vsi_queues_bind_intr(vsi,
+ !rte_intr_allow_others(intr_handle),
+ intr_handle->nb_efd,
+ intr_handle->intr_vec);
/* Enable interrupts for all the queues */
- ice_vsi_enable_queues_intr(vsi);
+ ice_vsi_enable_queues_intr(vsi, !rte_intr_allow_others(intr_handle));
rte_intr_enable(intr_handle);
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index da557a254..392b937a9 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -461,9 +461,12 @@ struct ice_vsi *
ice_setup_vsi(struct ice_pf *pf, enum ice_vsi_type type);
int
ice_release_vsi(struct ice_vsi *vsi);
-void ice_vsi_enable_queues_intr(struct ice_vsi *vsi);
-void ice_vsi_disable_queues_intr(struct ice_vsi *vsi);
-void ice_vsi_queues_bind_intr(struct ice_vsi *vsi);
+void ice_vsi_enable_queues_intr(struct ice_vsi *vsi, bool use_misc);
+void ice_vsi_disable_queues_intr(struct ice_vsi *vsi, bool use_misc);
+void ice_vsi_queues_bind_intr(struct ice_vsi *vsi,
+ bool use_misc,
+ int intr_num_max,
+ int *intr_vec);
static inline int
ice_align_floor(int n)
diff --git a/drivers/net/ice/ice_fdir_filter.c b/drivers/net/ice/ice_fdir_filter.c
index 5a791610f..533b4de5a 100644
--- a/drivers/net/ice/ice_fdir_filter.c
+++ b/drivers/net/ice/ice_fdir_filter.c
@@ -501,8 +501,8 @@ ice_fdir_setup(struct ice_pf *pf)
/* Enable FDIR MSIX interrupt */
vsi->nb_used_qps = 1;
- ice_vsi_queues_bind_intr(vsi);
- ice_vsi_enable_queues_intr(vsi);
+ ice_vsi_queues_bind_intr(vsi, false, 1, NULL);
+ ice_vsi_enable_queues_intr(vsi, true);
/* reserve memory for the fdir programming packet */
snprintf(z_name, sizeof(z_name), "ICE_%s_%d",
@@ -628,7 +628,7 @@ ice_fdir_teardown(struct ice_pf *pf)
if (!vsi)
return;
- ice_vsi_disable_queues_intr(vsi);
+ ice_vsi_disable_queues_intr(vsi, true);
err = ice_fdir_tx_queue_stop(eth_dev, pf->fdir.txq->queue_id);
if (err)
--
2.13.6
^ permalink raw reply [flat|nested] 3+ messages in thread* [dpdk-dev] [PATCH v3] net/ice: code clean for queue interrupt config
2020-03-15 3:22 [dpdk-dev] [PATCH] net/ice: fix wrong interrupt vector overwritten Qi Zhang
2020-03-15 7:40 ` [dpdk-dev] [PATCH v2] net/ice: code clean for queue interrupt config Qi Zhang
@ 2020-03-15 9:01 ` Qi Zhang
1 sibling, 0 replies; 3+ messages in thread
From: Qi Zhang @ 2020-03-15 9:01 UTC (permalink / raw)
To: beilei.xing, yahui.cao; +Cc: xiaolong.ye, dev, Qi Zhang
The patch decouple the intr_handle from ice_vsi_queue_bind_intr which
is shared by data Rx queue and fdir Rx queue's interrupt binding.
Though there is no issue in current implemenation of
ice_vsi_queue_bind_intr, but this rely on the assumption that the fdir
setup (when bind interrrupt for fdir queue) must be called before
dev_start (when bind interrupt for data queue), which is not a good
practise. More detail: if we deferred fdir setup to after dev_start,
it is possible when configure a fdir queue, a recorded vector for Rx
queue 0 be overwritten by the fdir queue's vector number, this may cause
interrupt Rx mode does not work on the Rx queue 0, since the vector
number be used by rte_eth_dev_rx_intr_ctl is corrupted.
The patch also apply the same code decoupling on
ice_vsi_enable|disable_queue_intr to make all queue interrupt API more
consistent.
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
--
v3:
- fix couple bugs
v2:
- rewrite the commit, actually this is not a fix, but code clean.
drivers/net/ice/ice_ethdev.c | 53 +++++++++++++++++----------------------
drivers/net/ice/ice_ethdev.h | 9 ++++---
drivers/net/ice/ice_fdir_filter.c | 6 ++---
3 files changed, 32 insertions(+), 36 deletions(-)
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index e59761c22..dd9105eed 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -2318,11 +2318,8 @@ ice_release_vsi(struct ice_vsi *vsi)
}
void
-ice_vsi_disable_queues_intr(struct ice_vsi *vsi)
+ice_vsi_disable_queues_intr(struct ice_vsi *vsi, bool use_misc)
{
- struct rte_eth_dev *dev = vsi->adapter->eth_dev;
- struct rte_pci_device *pci_dev = ICE_DEV_TO_PCI(dev);
- struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
struct ice_hw *hw = ICE_VSI_TO_HW(vsi);
uint16_t msix_intr, i;
@@ -2333,7 +2330,7 @@ ice_vsi_disable_queues_intr(struct ice_vsi *vsi)
rte_wmb();
}
- if (rte_intr_allow_others(intr_handle))
+ if (!use_misc)
/* vfio-pci */
for (i = 0; i < vsi->nb_msix; i++) {
msix_intr = vsi->msix_intr + i;
@@ -2368,7 +2365,8 @@ ice_dev_stop(struct rte_eth_dev *dev)
ice_tx_queue_stop(dev, i);
/* disable all queue interrupts */
- ice_vsi_disable_queues_intr(main_vsi);
+ ice_vsi_disable_queues_intr(main_vsi,
+ !rte_intr_allow_others(intr_handle));
if (pf->init_link_up)
ice_dev_set_link_up(dev);
@@ -2616,16 +2614,15 @@ __vsi_queues_bind_intr(struct ice_vsi *vsi, uint16_t msix_vect,
}
void
-ice_vsi_queues_bind_intr(struct ice_vsi *vsi)
+ice_vsi_queues_bind_intr(struct ice_vsi *vsi,
+ bool use_misc,
+ int intr_num_max,
+ int *intr_vec)
{
- struct rte_eth_dev *dev = vsi->adapter->eth_dev;
- struct rte_pci_device *pci_dev = ICE_DEV_TO_PCI(dev);
- struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
struct ice_hw *hw = ICE_VSI_TO_HW(vsi);
uint16_t msix_vect = vsi->msix_intr;
- uint16_t nb_msix = RTE_MIN(vsi->nb_msix, intr_handle->nb_efd);
+ uint16_t nb_msix = RTE_MIN(vsi->nb_msix, intr_num_max);
uint16_t queue_idx = 0;
- int record = 0;
int i;
/* clear Rx/Tx queue interrupt */
@@ -2634,15 +2631,12 @@ ice_vsi_queues_bind_intr(struct ice_vsi *vsi)
ICE_WRITE_REG(hw, QINT_RQCTL(vsi->base_queue + i), 0);
}
- /* PF bind interrupt */
- if (rte_intr_dp_is_en(intr_handle)) {
- queue_idx = 0;
- record = 1;
- }
+ if (use_misc)
+ nb_msix = 1;
for (i = 0; i < vsi->nb_used_qps; i++) {
if (nb_msix <= 1) {
- if (!rte_intr_allow_others(intr_handle))
+ if (use_misc)
msix_vect = ICE_MISC_VEC_ID;
/* uio mapping all queue to one msix_vect */
@@ -2650,9 +2644,8 @@ ice_vsi_queues_bind_intr(struct ice_vsi *vsi)
vsi->base_queue + i,
vsi->nb_used_qps - i);
- for (; !!record && i < vsi->nb_used_qps; i++)
- intr_handle->intr_vec[queue_idx + i] =
- msix_vect;
+ for (; intr_vec && i < vsi->nb_used_qps; i++)
+ intr_vec[queue_idx + i] = msix_vect;
break;
}
@@ -2660,8 +2653,8 @@ ice_vsi_queues_bind_intr(struct ice_vsi *vsi)
__vsi_queues_bind_intr(vsi, msix_vect,
vsi->base_queue + i, 1);
- if (!!record)
- intr_handle->intr_vec[queue_idx + i] = msix_vect;
+ if (intr_vec)
+ intr_vec[queue_idx + i] = msix_vect;
msix_vect++;
nb_msix--;
@@ -2669,15 +2662,12 @@ ice_vsi_queues_bind_intr(struct ice_vsi *vsi)
}
void
-ice_vsi_enable_queues_intr(struct ice_vsi *vsi)
+ice_vsi_enable_queues_intr(struct ice_vsi *vsi, bool use_misc)
{
- struct rte_eth_dev *dev = vsi->adapter->eth_dev;
- struct rte_pci_device *pci_dev = ICE_DEV_TO_PCI(dev);
- struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
struct ice_hw *hw = ICE_VSI_TO_HW(vsi);
uint16_t msix_intr, i;
- if (rte_intr_allow_others(intr_handle))
+ if (!use_misc)
for (i = 0; i < vsi->nb_used_qps; i++) {
msix_intr = vsi->msix_intr + i;
ICE_WRITE_REG(hw, GLINT_DYN_CTL(msix_intr),
@@ -2733,10 +2723,13 @@ ice_rxq_intr_setup(struct rte_eth_dev *dev)
/* Map queues with MSIX interrupt */
vsi->nb_used_qps = dev->data->nb_rx_queues;
- ice_vsi_queues_bind_intr(vsi);
+ ice_vsi_queues_bind_intr(vsi,
+ !rte_intr_allow_others(intr_handle),
+ intr_handle->nb_efd,
+ intr_handle->intr_vec);
/* Enable interrupts for all the queues */
- ice_vsi_enable_queues_intr(vsi);
+ ice_vsi_enable_queues_intr(vsi, !rte_intr_allow_others(intr_handle));
rte_intr_enable(intr_handle);
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index da557a254..392b937a9 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -461,9 +461,12 @@ struct ice_vsi *
ice_setup_vsi(struct ice_pf *pf, enum ice_vsi_type type);
int
ice_release_vsi(struct ice_vsi *vsi);
-void ice_vsi_enable_queues_intr(struct ice_vsi *vsi);
-void ice_vsi_disable_queues_intr(struct ice_vsi *vsi);
-void ice_vsi_queues_bind_intr(struct ice_vsi *vsi);
+void ice_vsi_enable_queues_intr(struct ice_vsi *vsi, bool use_misc);
+void ice_vsi_disable_queues_intr(struct ice_vsi *vsi, bool use_misc);
+void ice_vsi_queues_bind_intr(struct ice_vsi *vsi,
+ bool use_misc,
+ int intr_num_max,
+ int *intr_vec);
static inline int
ice_align_floor(int n)
diff --git a/drivers/net/ice/ice_fdir_filter.c b/drivers/net/ice/ice_fdir_filter.c
index 6342b560c..fdcfbd78f 100644
--- a/drivers/net/ice/ice_fdir_filter.c
+++ b/drivers/net/ice/ice_fdir_filter.c
@@ -501,8 +501,8 @@ ice_fdir_setup(struct ice_pf *pf)
/* Enable FDIR MSIX interrupt */
vsi->nb_used_qps = 1;
- ice_vsi_queues_bind_intr(vsi);
- ice_vsi_enable_queues_intr(vsi);
+ ice_vsi_queues_bind_intr(vsi, true, 1, NULL);
+ ice_vsi_enable_queues_intr(vsi, true);
/* reserve memory for the fdir programming packet */
snprintf(z_name, sizeof(z_name), "ICE_%s_%d",
@@ -628,7 +628,7 @@ ice_fdir_teardown(struct ice_pf *pf)
if (!vsi)
return;
- ice_vsi_disable_queues_intr(vsi);
+ ice_vsi_disable_queues_intr(vsi, true);
err = ice_fdir_tx_queue_stop(eth_dev, pf->fdir.txq->queue_id);
if (err)
--
2.13.6
^ permalink raw reply [flat|nested] 3+ messages in thread