[v3] net/ice: fix DCF state checking mechanism
Checks
Commit Message
From: Peng Zhang <peng1x.zhang@intel.com>
DCF state previously checking mechanism cannot detect DCF state
correctly in given situation,so PMD will report incorrect error
code and mislead user. In this patch, DCF state checking mechanism
is fixed through create or delete rule successfully or not. Then
error code will be set as EAGAIN in given situation, while not EINVAL.
The given situation as following steps show:
step 1. DPDK DCF state has been set to on after VF has reset and multiple
rules are creating.
step 2. A VF reset happen immediately, kernel send an event to DPDK DCF
and set STATE to pause.
step 3. Before DPDK DCF receive the event, it is possible a rule creation
is ongoing,then in virtual channel queue, the rule request is in front of
the "re-connect", then it will be rejected.
step 4. But the DPDK DCF state is not set to pause, error code will be set
as EINVAL, not EAGAIN.
Fixes: 285f63fc6bb7 ("net/ice: track DCF state of PF")
Cc: stable@dpdk.org
Signed-off-by: Peng Zhang <peng1x.zhang@intel.com>
---
v3 Changes:
- Add the situation description, expected error code and incorrect error code in commit log.
v2 Changes:
- Modify DCF state checking mechanism.
drivers/net/ice/ice_dcf_parent.c | 3 ---
drivers/net/ice/ice_switch_filter.c | 20 ++++++--------------
2 files changed, 6 insertions(+), 17 deletions(-)
Comments
> -----Original Message-----
> From: Zhang, Peng1X <peng1x.zhang@intel.com>
> Sent: Saturday, May 21, 2022 2:31 AM
> To: Yang, Qiming <qiming.yang@intel.com>; Zhang, Qi Z
> <qi.z.zhang@intel.com>; dev@dpdk.org
> Cc: Zhang, Peng1X <peng1x.zhang@intel.com>; stable@dpdk.org
> Subject: [PATCH v3] net/ice: fix DCF state checking mechanism
>
> From: Peng Zhang <peng1x.zhang@intel.com>
>
> DCF state previously checking mechanism cannot detect DCF state correctly in
> given situation,so PMD will report incorrect error code and mislead user. In this
> patch, DCF state checking mechanism is fixed through create or delete rule
> successfully or not. Then error code will be set as EAGAIN in given situation,
> while not EINVAL.
We should not assume the error code is always EAGAIN if any rule is rejected by the kernel PF driver, this will also mislead the application.
We need the kernel driver to return a clear error code, then DCF pass them to the application.
@@ -121,7 +121,6 @@ ice_dcf_vsi_update_service_handler(void *param)
struct ice_dcf_hw *hw = reset_param->dcf_hw;
struct ice_dcf_adapter *adapter =
container_of(hw, struct ice_dcf_adapter, real_hw);
- struct ice_adapter *parent_adapter = &adapter->parent;
pthread_detach(pthread_self());
@@ -130,8 +129,6 @@ ice_dcf_vsi_update_service_handler(void *param)
rte_spinlock_lock(&vsi_update_lock);
if (!ice_dcf_handle_vsi_update_event(hw)) {
- __atomic_store_n(&parent_adapter->dcf_state_on, true,
- __ATOMIC_RELAXED);
ice_dcf_update_vf_vsi_map(&adapter->parent.hw,
hw->num_vfs, hw->vf_vsi_map);
}
@@ -403,13 +403,6 @@ ice_switch_create(struct ice_adapter *ad,
goto error;
}
- if (ice_dcf_adminq_need_retry(ad)) {
- rte_flow_error_set(error, EAGAIN,
- RTE_FLOW_ERROR_TYPE_ITEM, NULL,
- "DCF is not on");
- goto error;
- }
-
ret = ice_add_adv_rule(hw, list, lkups_cnt, rule_info, &rule_added);
if (!ret) {
filter_conf_ptr = rte_zmalloc("ice_switch_filter",
@@ -432,6 +425,9 @@ ice_switch_create(struct ice_adapter *ad,
filter_conf_ptr->fltr_status = ICE_SW_FLTR_ADDED;
flow->rule = filter_conf_ptr;
+
+ if (ad->hw.dcf_enabled)
+ __atomic_store_n(&ad->dcf_state_on, true, __ATOMIC_RELAXED);
} else {
if (ice_dcf_adminq_need_retry(ad))
ret = -EAGAIN;
@@ -490,13 +486,6 @@ ice_switch_destroy(struct ice_adapter *ad,
return -rte_errno;
}
- if (ice_dcf_adminq_need_retry(ad)) {
- rte_flow_error_set(error, EAGAIN,
- RTE_FLOW_ERROR_TYPE_ITEM, NULL,
- "DCF is not on");
- return -rte_errno;
- }
-
ret = ice_rem_adv_rule_by_id(hw, &filter_conf_ptr->sw_query_data);
if (ret) {
if (ice_dcf_adminq_need_retry(ad))
@@ -508,6 +497,9 @@ ice_switch_destroy(struct ice_adapter *ad,
RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
"fail to destroy switch filter rule");
return -rte_errno;
+ } else {
+ if (ad->hw.dcf_enabled)
+ __atomic_store_n(&ad->dcf_state_on, true, __ATOMIC_RELAXED);
}
ice_switch_filter_rule_free(flow);