@@ -226,7 +226,6 @@ enum ice_pf_state {
ICE_VFLR_EVENT_PENDING,
ICE_FLTR_OVERFLOW_PROMISC,
ICE_VF_DIS,
- ICE_VF_DEINIT_IN_PROGRESS,
ICE_CFG_BUSY,
ICE_SERVICE_SCHED,
ICE_SERVICE_DIS,
@@ -1597,7 +1597,9 @@ static void ice_handle_mdd_event(struct ice_pf *pf)
* reset, so print the event prior to reset.
*/
ice_print_vf_rx_mdd_event(vf);
+ mutex_lock(&pf->vf[i].cfg_lock);
ice_reset_vf(&pf->vf[i], false);
+ mutex_unlock(&pf->vf[i].cfg_lock);
}
}
}
@@ -615,8 +615,6 @@ void ice_free_vfs(struct ice_pf *pf)
struct ice_hw *hw = &pf->hw;
unsigned int tmp, i;
- set_bit(ICE_VF_DEINIT_IN_PROGRESS, pf->state);
-
if (!pf->vf)
return;
@@ -632,22 +630,26 @@ void ice_free_vfs(struct ice_pf *pf)
else
dev_warn(dev, "VFs are assigned - not disabling SR-IOV\n");
- /* Avoid wait time by stopping all VFs at the same time */
- ice_for_each_vf(pf, i)
- ice_dis_vf_qs(&pf->vf[i]);
-
tmp = pf->num_alloc_vfs;
pf->num_qps_per_vf = 0;
pf->num_alloc_vfs = 0;
for (i = 0; i < tmp; i++) {
- if (test_bit(ICE_VF_STATE_INIT, pf->vf[i].vf_states)) {
+ struct ice_vf *vf = &pf->vf[i];
+
+ mutex_lock(&vf->cfg_lock);
+
+ ice_dis_vf_qs(vf);
+
+ if (test_bit(ICE_VF_STATE_INIT, vf->vf_states)) {
/* disable VF qp mappings and set VF disable state */
- ice_dis_vf_mappings(&pf->vf[i]);
- set_bit(ICE_VF_STATE_DIS, pf->vf[i].vf_states);
- ice_free_vf_res(&pf->vf[i]);
+ ice_dis_vf_mappings(vf);
+ set_bit(ICE_VF_STATE_DIS, vf->vf_states);
+ ice_free_vf_res(vf);
}
- mutex_destroy(&pf->vf[i].cfg_lock);
+ mutex_unlock(&vf->cfg_lock);
+
+ mutex_destroy(&vf->cfg_lock);
}
if (ice_sriov_free_msix_res(pf))
@@ -683,7 +685,6 @@ void ice_free_vfs(struct ice_pf *pf)
i);
clear_bit(ICE_VF_DIS, pf->state);
- clear_bit(ICE_VF_DEINIT_IN_PROGRESS, pf->state);
clear_bit(ICE_FLAG_SRIOV_ENA, pf->flags);
}
@@ -1565,6 +1566,8 @@ bool ice_reset_all_vfs(struct ice_pf *pf, bool is_vflr)
ice_for_each_vf(pf, v) {
vf = &pf->vf[v];
+ mutex_lock(&vf->cfg_lock);
+
vf->driver_caps = 0;
ice_vc_set_default_allowlist(vf);
@@ -1578,6 +1581,8 @@ bool ice_reset_all_vfs(struct ice_pf *pf, bool is_vflr)
ice_vf_pre_vsi_rebuild(vf);
ice_vf_rebuild_vsi(vf);
ice_vf_post_vsi_rebuild(vf);
+
+ mutex_unlock(&vf->cfg_lock);
}
ice_flush(hw);
@@ -1624,6 +1629,8 @@ bool ice_reset_vf(struct ice_vf *vf, bool is_vflr)
u32 reg;
int i;
+ lockdep_assert_held(&vf->cfg_lock);
+
dev = ice_pf_to_dev(pf);
if (test_bit(ICE_VF_RESETS_DISABLED, pf->state)) {
@@ -2110,9 +2117,12 @@ void ice_process_vflr_event(struct ice_pf *pf)
bit_idx = (hw->func_caps.vf_base_id + vf_id) % 32;
/* read GLGEN_VFLRSTAT register to find out the flr VFs */
reg = rd32(hw, GLGEN_VFLRSTAT(reg_idx));
- if (reg & BIT(bit_idx))
+ if (reg & BIT(bit_idx)) {
/* GLGEN_VFLRSTAT bit will be cleared in ice_reset_vf */
+ mutex_lock(&vf->cfg_lock);
ice_reset_vf(vf, true);
+ mutex_unlock(&vf->cfg_lock);
+ }
}
}
@@ -2189,7 +2199,9 @@ ice_vf_lan_overflow_event(struct ice_pf *pf, struct ice_rq_event_info *event)
if (!vf)
return;
+ mutex_lock(&vf->cfg_lock);
ice_vc_reset_vf(vf);
+ mutex_unlock(&vf->cfg_lock);
}
/**
@@ -4300,10 +4312,6 @@ void ice_vc_process_vf_msg(struct ice_pf *pf, struct ice_rq_event_info *event)
struct device *dev;
int err = 0;
- /* if de-init is underway, don't process messages from VF */
- if (test_bit(ICE_VF_DEINIT_IN_PROGRESS, pf->state))
- return;
-
dev = ice_pf_to_dev(pf);
if (ice_validate_vf_id(pf, vf_id)) {
err = -EINVAL;
commit fadead80fe4c033b5e514fcbadd20b55c4494112 upstream. Commit c503e63200c6 ("ice: Stop processing VF messages during teardown") introduced a driver state flag, ICE_VF_DEINIT_IN_PROGRESS, which is intended to prevent some issues with concurrently handling messages from VFs while tearing down the VFs. This change was motivated by crashes caused while tearing down and bringing up VFs in rapid succession. It turns out that the fix actually introduces issues with the VF driver caused because the PF no longer responds to any messages sent by the VF during its .remove routine. This results in the VF potentially removing its DMA memory before the PF has shut down the device queues. Additionally, the fix doesn't actually resolve concurrency issues within the ice driver. It is possible for a VF to initiate a reset just prior to the ice driver removing VFs. This can result in the remove task concurrently operating while the VF is being reset. This results in similar memory corruption and panics purportedly fixed by that commit. Fix this concurrency at its root by protecting both the reset and removal flows using the existing VF cfg_lock. This ensures that we cannot remove the VF while any outstanding critical tasks such as a virtchnl message or a reset are occurring. This locking change also fixes the root cause originally fixed by commit c503e63200c6 ("ice: Stop processing VF messages during teardown"), so we can simply revert it. Note that I kept these two changes together because simply reverting the original commit alone would leave the driver vulnerable to worse race conditions. Fixes: c503e63200c6 ("ice: Stop processing VF messages during teardown") Cc: <stable@vger.kernel.org> # 5.13.x: e6ba5273d4ed: ice: Fix race conditions between virtchnl handling and VF ndo ops Cc: <stable@vger.kernel.org> # 5.13.x: b385cca47363: ice: Fix not stopping Tx queues for VF Cc: <stable@vger.kernel.org> # 5.13.x Signed-off-by: Jacob Keller <jacob.e.keller@intel.com> Tested-by: Konrad Jankowski <konrad0.jankowski@intel.com> Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> --- This should apply to 5.13.x drivers/net/ethernet/intel/ice/ice.h | 1 - drivers/net/ethernet/intel/ice/ice_main.c | 2 + .../net/ethernet/intel/ice/ice_virtchnl_pf.c | 42 +++++++++++-------- 3 files changed, 27 insertions(+), 18 deletions(-)