summaryrefslogtreecommitdiff
path: root/drivers/net/ethernet/ibm/ibmvnic.c
diff options
context:
space:
mode:
authorJason Liu <jason.hui.liu@nxp.com>2022-09-26 12:56:22 -0500
committerJason Liu <jason.hui.liu@nxp.com>2022-09-26 12:56:22 -0500
commit2b53d4d52360709aebc699b18fde867a571b8c39 (patch)
treec2ff72465262fbd327e6e09361154c52047ef792 /drivers/net/ethernet/ibm/ibmvnic.c
parent36363d8623ba60858e2632b7d2b70dae932c9a8b (diff)
parent3e98e33d345e981800e03dd29f6f6343286d30b6 (diff)
Merge tag 'v5.15.70' into lf-5.15.y
This is the 5.15.70 stable release * tag 'v5.15.70': (2444 commits) Linux 5.15.70 ALSA: hda/sigmatel: Fix unused variable warning for beep power change cgroup: Add missing cpus_read_lock() to cgroup_attach_task_all() ... Signed-off-by: Jason Liu <jason.hui.liu@nxp.com> Conflicts: arch/arm/boot/dts/imx6ul.dtsi arch/arm/mm/mmu.c arch/arm64/boot/dts/freescale/imx8mp-evk.dts drivers/gpu/drm/imx/dcss/dcss-kms.c drivers/media/platform/nxp/imx-jpeg/mxc-jpeg.c drivers/media/platform/nxp/imx-jpeg/mxc-jpeg.h drivers/mtd/nand/raw/gpmi-nand/gpmi-nand.c drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c drivers/soc/fsl/Kconfig drivers/soc/imx/gpcv2.c drivers/usb/dwc3/host.c net/dsa/slave.c sound/soc/fsl/imx-card.c
Diffstat (limited to 'drivers/net/ethernet/ibm/ibmvnic.c')
-rw-r--r--drivers/net/ethernet/ibm/ibmvnic.c147
1 files changed, 132 insertions, 15 deletions
diff --git a/drivers/net/ethernet/ibm/ibmvnic.c b/drivers/net/ethernet/ibm/ibmvnic.c
index b262aa84b6a2..4a070724a8fb 100644
--- a/drivers/net/ethernet/ibm/ibmvnic.c
+++ b/drivers/net/ethernet/ibm/ibmvnic.c
@@ -2064,6 +2064,19 @@ static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
}
/*
+ * Initialize the init_done completion and return code values. We
+ * can get a transport event just after registering the CRQ and the
+ * tasklet will use this to communicate the transport event. To ensure
+ * we don't miss the notification/error, initialize these _before_
+ * regisering the CRQ.
+ */
+static inline void reinit_init_done(struct ibmvnic_adapter *adapter)
+{
+ reinit_completion(&adapter->init_done);
+ adapter->init_done_rc = 0;
+}
+
+/*
* do_reset returns zero if we are able to keep processing reset events, or
* non-zero if we hit a fatal error and must halt.
*/
@@ -2169,6 +2182,8 @@ static int do_reset(struct ibmvnic_adapter *adapter,
*/
adapter->state = VNIC_PROBED;
+ reinit_init_done(adapter);
+
if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
rc = init_crq_queue(adapter);
} else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
@@ -2314,7 +2329,8 @@ static int do_hard_reset(struct ibmvnic_adapter *adapter,
*/
adapter->state = VNIC_PROBED;
- reinit_completion(&adapter->init_done);
+ reinit_init_done(adapter);
+
rc = init_crq_queue(adapter);
if (rc) {
netdev_err(adapter->netdev,
@@ -2455,23 +2471,82 @@ out:
static void __ibmvnic_reset(struct work_struct *work)
{
struct ibmvnic_adapter *adapter;
- bool saved_state = false;
+ unsigned int timeout = 5000;
struct ibmvnic_rwi *tmprwi;
+ bool saved_state = false;
struct ibmvnic_rwi *rwi;
unsigned long flags;
- u32 reset_state;
+ struct device *dev;
+ bool need_reset;
int num_fails = 0;
+ u32 reset_state;
int rc = 0;
adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
+ dev = &adapter->vdev->dev;
- if (test_and_set_bit_lock(0, &adapter->resetting)) {
+ /* Wait for ibmvnic_probe() to complete. If probe is taking too long
+ * or if another reset is in progress, defer work for now. If probe
+ * eventually fails it will flush and terminate our work.
+ *
+ * Three possibilities here:
+ * 1. Adpater being removed - just return
+ * 2. Timed out on probe or another reset in progress - delay the work
+ * 3. Completed probe - perform any resets in queue
+ */
+ if (adapter->state == VNIC_PROBING &&
+ !wait_for_completion_timeout(&adapter->probe_done, timeout)) {
+ dev_err(dev, "Reset thread timed out on probe");
queue_delayed_work(system_long_wq,
&adapter->ibmvnic_delayed_reset,
IBMVNIC_RESET_DELAY);
return;
}
+ /* adapter is done with probe (i.e state is never VNIC_PROBING now) */
+ if (adapter->state == VNIC_REMOVING)
+ return;
+
+ /* ->rwi_list is stable now (no one else is removing entries) */
+
+ /* ibmvnic_probe() may have purged the reset queue after we were
+ * scheduled to process a reset so there maybe no resets to process.
+ * Before setting the ->resetting bit though, we have to make sure
+ * that there is infact a reset to process. Otherwise we may race
+ * with ibmvnic_open() and end up leaving the vnic down:
+ *
+ * __ibmvnic_reset() ibmvnic_open()
+ * ----------------- --------------
+ *
+ * set ->resetting bit
+ * find ->resetting bit is set
+ * set ->state to IBMVNIC_OPEN (i.e
+ * assume reset will open device)
+ * return
+ * find reset queue empty
+ * return
+ *
+ * Neither performed vnic login/open and vnic stays down
+ *
+ * If we hold the lock and conditionally set the bit, either we
+ * or ibmvnic_open() will complete the open.
+ */
+ need_reset = false;
+ spin_lock(&adapter->rwi_lock);
+ if (!list_empty(&adapter->rwi_list)) {
+ if (test_and_set_bit_lock(0, &adapter->resetting)) {
+ queue_delayed_work(system_long_wq,
+ &adapter->ibmvnic_delayed_reset,
+ IBMVNIC_RESET_DELAY);
+ } else {
+ need_reset = true;
+ }
+ }
+ spin_unlock(&adapter->rwi_lock);
+
+ if (!need_reset)
+ return;
+
rwi = get_next_rwi(adapter);
while (rwi) {
spin_lock_irqsave(&adapter->state_lock, flags);
@@ -2623,13 +2698,6 @@ static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
goto err;
}
- if (adapter->state == VNIC_PROBING) {
- netdev_warn(netdev, "Adapter reset during probe\n");
- adapter->init_done_rc = -EAGAIN;
- ret = EAGAIN;
- goto err;
- }
-
list_for_each_entry(tmp, &adapter->rwi_list, list) {
if (tmp->reset_reason == reason) {
netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
@@ -5485,10 +5553,6 @@ static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
adapter->from_passive_init = false;
- if (reset)
- reinit_completion(&adapter->init_done);
-
- adapter->init_done_rc = 0;
rc = ibmvnic_send_crq_init(adapter);
if (rc) {
dev_err(dev, "Send crq init failed with error %d\n", rc);
@@ -5502,12 +5566,14 @@ static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
if (adapter->init_done_rc) {
release_crq_queue(adapter);
+ dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc);
return adapter->init_done_rc;
}
if (adapter->from_passive_init) {
adapter->state = VNIC_OPEN;
adapter->from_passive_init = false;
+ dev_err(dev, "CRQ-init failed, passive-init\n");
return -1;
}
@@ -5519,6 +5585,15 @@ static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
release_sub_crqs(adapter, 0);
rc = init_sub_crqs(adapter);
} else {
+ /* no need to reinitialize completely, but we do
+ * need to clean up transmits that were in flight
+ * when we processed the reset. Failure to do so
+ * will confound the upper layer, usually TCP, by
+ * creating the illusion of transmits that are
+ * awaiting completion.
+ */
+ clean_tx_pools(adapter);
+
rc = reset_sub_crq_queues(adapter);
}
} else {
@@ -5547,6 +5622,7 @@ static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
struct ibmvnic_adapter *adapter;
struct net_device *netdev;
unsigned char *mac_addr_p;
+ unsigned long flags;
bool init_success;
int rc;
@@ -5588,6 +5664,7 @@ static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
spin_lock_init(&adapter->rwi_lock);
spin_lock_init(&adapter->state_lock);
mutex_init(&adapter->fw_lock);
+ init_completion(&adapter->probe_done);
init_completion(&adapter->init_done);
init_completion(&adapter->fw_done);
init_completion(&adapter->reset_done);
@@ -5596,6 +5673,33 @@ static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
init_success = false;
do {
+ reinit_init_done(adapter);
+
+ /* clear any failovers we got in the previous pass
+ * since we are reinitializing the CRQ
+ */
+ adapter->failover_pending = false;
+
+ /* If we had already initialized CRQ, we may have one or
+ * more resets queued already. Discard those and release
+ * the CRQ before initializing the CRQ again.
+ */
+ release_crq_queue(adapter);
+
+ /* Since we are still in PROBING state, __ibmvnic_reset()
+ * will not access the ->rwi_list and since we released CRQ,
+ * we won't get _new_ transport events. But there maybe an
+ * ongoing ibmvnic_reset() call. So serialize access to
+ * rwi_list. If we win the race, ibvmnic_reset() could add
+ * a reset after we purged but thats ok - we just may end
+ * up with an extra reset (i.e similar to having two or more
+ * resets in the queue at once).
+ * CHECK.
+ */
+ spin_lock_irqsave(&adapter->rwi_lock, flags);
+ flush_reset_queue(adapter);
+ spin_unlock_irqrestore(&adapter->rwi_lock, flags);
+
rc = init_crq_queue(adapter);
if (rc) {
dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
@@ -5647,6 +5751,8 @@ static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
}
dev_info(&dev->dev, "ibmvnic registered\n");
+ complete(&adapter->probe_done);
+
return 0;
ibmvnic_register_fail:
@@ -5661,6 +5767,17 @@ ibmvnic_stats_fail:
ibmvnic_init_fail:
release_sub_crqs(adapter, 1);
release_crq_queue(adapter);
+
+ /* cleanup worker thread after releasing CRQ so we don't get
+ * transport events (i.e new work items for the worker thread).
+ */
+ adapter->state = VNIC_REMOVING;
+ complete(&adapter->probe_done);
+ flush_work(&adapter->ibmvnic_reset);
+ flush_delayed_work(&adapter->ibmvnic_delayed_reset);
+
+ flush_reset_queue(adapter);
+
mutex_destroy(&adapter->fw_lock);
free_netdev(netdev);