[dpdk-dev] [PATCH v2] net/vmxnet3: add interrupt support in MSI-X
Jochen Behrens
jbehrens at vmware.com
Thu Jul 8 16:02:25 CEST 2021
Add support for MSI-X interrupt vectors to the vmxnet3 driver.
This will allow more efficient deployments in cloud environments.
By default it will try to allocate 1 vector (0) for link
event and one MSI-X vector for each Rx queue. To simplify
things, it will only be enabled if the number of Tx and Rx
queues are equal (so that Tx/Rx share the same vector).
If for any reason vmxnet3 cannot enable intr mode, it will
fall back to the LSC only mode.
v2: addressed formatting nits in comments
Signed-off-by: Yong Wang <yongwang at vmware.com>
Signed-off-by: Jochen Behrens <jbehrens at vmware.com>
---
drivers/net/vmxnet3/vmxnet3_ethdev.c | 340 +++++++++++++++++++++------
drivers/net/vmxnet3/vmxnet3_ethdev.h | 12 +
2 files changed, 284 insertions(+), 68 deletions(-)
diff --git a/drivers/net/vmxnet3/vmxnet3_ethdev.c b/drivers/net/vmxnet3/vmxnet3_ethdev.c
index 5bffbb8a0e..1a3291273a 100644
--- a/drivers/net/vmxnet3/vmxnet3_ethdev.c
+++ b/drivers/net/vmxnet3/vmxnet3_ethdev.c
@@ -38,8 +38,6 @@
#include "vmxnet3_logs.h"
#include "vmxnet3_ethdev.h"
-#define PROCESS_SYS_EVENTS 0
-
#define VMXNET3_TX_MAX_SEG UINT8_MAX
#define VMXNET3_TX_OFFLOAD_CAP \
@@ -96,7 +94,12 @@ static int vmxnet3_dev_vlan_filter_set(struct rte_eth_dev *dev,
static int vmxnet3_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
static int vmxnet3_mac_addr_set(struct rte_eth_dev *dev,
struct rte_ether_addr *mac_addr);
+static void vmxnet3_process_events(struct rte_eth_dev *dev);
static void vmxnet3_interrupt_handler(void *param);
+static int vmxnet3_dev_rx_queue_intr_enable(struct rte_eth_dev *dev,
+ uint16_t queue_id);
+static int vmxnet3_dev_rx_queue_intr_disable(struct rte_eth_dev *dev,
+ uint16_t queue_id);
/*
* The set of PCI devices this driver supports
@@ -133,6 +136,8 @@ static const struct eth_dev_ops vmxnet3_eth_dev_ops = {
.rx_queue_release = vmxnet3_dev_rx_queue_release,
.tx_queue_setup = vmxnet3_dev_tx_queue_setup,
.tx_queue_release = vmxnet3_dev_tx_queue_release,
+ .rx_queue_intr_enable = vmxnet3_dev_rx_queue_intr_enable,
+ .rx_queue_intr_disable = vmxnet3_dev_rx_queue_intr_disable,
};
struct vmxnet3_xstats_name_off {
@@ -183,30 +188,61 @@ gpa_zone_reserve(struct rte_eth_dev *dev, uint32_t size,
}
/*
- * This function is based on vmxnet3_disable_intr()
+ * Enable the given interrupt
*/
static void
-vmxnet3_disable_intr(struct vmxnet3_hw *hw)
+vmxnet3_enable_intr(struct vmxnet3_hw *hw, unsigned int intr_idx)
{
- int i;
+ PMD_INIT_FUNC_TRACE();
+ VMXNET3_WRITE_BAR0_REG(hw, VMXNET3_REG_IMR + intr_idx * 8, 0);
+}
+/*
+ * Disable the given interrupt
+ */
+static void
+vmxnet3_disable_intr(struct vmxnet3_hw *hw, unsigned int intr_idx)
+{
PMD_INIT_FUNC_TRACE();
+ VMXNET3_WRITE_BAR0_REG(hw, VMXNET3_REG_IMR + intr_idx * 8, 1);
+}
- hw->shared->devRead.intrConf.intrCtrl |= VMXNET3_IC_DISABLE_ALL;
- for (i = 0; i < hw->num_intrs; i++)
- VMXNET3_WRITE_BAR0_REG(hw, VMXNET3_REG_IMR + i * 8, 1);
+/*
+ * Enable all intrs used by the device
+ */
+static void
+vmxnet3_enable_all_intrs(struct vmxnet3_hw *hw)
+{
+ Vmxnet3_DSDevRead *devRead = &hw->shared->devRead;
+
+ PMD_INIT_FUNC_TRACE();
+
+ devRead->intrConf.intrCtrl &= rte_cpu_to_le_32(~VMXNET3_IC_DISABLE_ALL);
+
+ if (hw->intr.lsc_only) {
+ vmxnet3_enable_intr(hw, devRead->intrConf.eventIntrIdx);
+ } else {
+ int i;
+
+ for (i = 0; i < hw->intr.num_intrs; i++)
+ vmxnet3_enable_intr(hw, i);
+ }
}
+/*
+ * Disable all intrs used by the device
+ */
static void
-vmxnet3_enable_intr(struct vmxnet3_hw *hw)
+vmxnet3_disable_all_intrs(struct vmxnet3_hw *hw)
{
int i;
PMD_INIT_FUNC_TRACE();
- hw->shared->devRead.intrConf.intrCtrl &= ~VMXNET3_IC_DISABLE_ALL;
+ hw->shared->devRead.intrConf.intrCtrl |=
+ rte_cpu_to_le_32(VMXNET3_IC_DISABLE_ALL);
for (i = 0; i < hw->num_intrs; i++)
- VMXNET3_WRITE_BAR0_REG(hw, VMXNET3_REG_IMR + i * 8, 0);
+ vmxnet3_disable_intr(hw, i);
}
/*
@@ -304,7 +340,7 @@ eth_vmxnet3_dev_init(struct rte_eth_dev *eth_dev)
return -EIO;
}
- PMD_INIT_LOG(DEBUG, "Using device version %d\n", hw->version);
+ PMD_INIT_LOG(INFO, "Using device v%d", hw->version);
/* Check UPT version compatibility with driver. */
ver = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_UVRS);
@@ -408,6 +444,40 @@ static struct rte_pci_driver rte_vmxnet3_pmd = {
.remove = eth_vmxnet3_pci_remove,
};
+static void
+vmxnet3_alloc_intr_resources(struct rte_eth_dev *dev)
+{
+ struct vmxnet3_hw *hw = dev->data->dev_private;
+ uint32_t cfg;
+ int nvec = 1; /* for link event */
+
+ /* intr settings */
+ VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
+ VMXNET3_CMD_GET_CONF_INTR);
+ cfg = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
+ hw->intr.type = cfg & 0x3;
+ hw->intr.mask_mode = (cfg >> 2) & 0x3;
+
+ if (hw->intr.type == VMXNET3_IT_AUTO)
+ hw->intr.type = VMXNET3_IT_MSIX;
+
+ if (hw->intr.type == VMXNET3_IT_MSIX) {
+ /* only support shared tx/rx intr */
+ if (hw->num_tx_queues != hw->num_rx_queues)
+ goto msix_err;
+
+ nvec += hw->num_rx_queues;
+ hw->intr.num_intrs = nvec;
+ return;
+ }
+
+msix_err:
+ /* the tx/rx queue interrupt will be disabled */
+ hw->intr.num_intrs = 2;
+ hw->intr.lsc_only = TRUE;
+ PMD_INIT_LOG(INFO, "Enabled MSI-X with %d vectors", hw->intr.num_intrs);
+}
+
static int
vmxnet3_dev_configure(struct rte_eth_dev *dev)
{
@@ -494,6 +564,8 @@ vmxnet3_dev_configure(struct rte_eth_dev *dev)
hw->rss_confPA = mz->iova;
}
+ vmxnet3_alloc_intr_resources(dev);
+
return 0;
}
@@ -514,6 +586,85 @@ vmxnet3_write_mac(struct vmxnet3_hw *hw, const uint8_t *addr)
VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_MACH, val);
}
+/*
+ * Configure the hardware to generate MSI-X interrupts.
+ * If setting up MSIx fails, try setting up MSI (only 1 interrupt vector
+ * which will be disabled to allow lsc to work).
+ *
+ * Returns 0 on success and -1 otherwise.
+ */
+static int
+vmxnet3_configure_msix(struct rte_eth_dev *dev)
+{
+ struct vmxnet3_hw *hw = dev->data->dev_private;
+ struct rte_intr_handle *intr_handle = dev->intr_handle;
+ uint16_t intr_vector;
+ int i;
+
+ hw->intr.event_intr_idx = 0;
+
+ /* only vfio-pci driver can support interrupt mode. */
+ if (!rte_intr_cap_multiple(intr_handle) ||
+ dev->data->dev_conf.intr_conf.rxq == 0)
+ return -1;
+
+ intr_vector = dev->data->nb_rx_queues;
+ if (intr_vector > VMXNET3_MAX_RX_QUEUES) {
+ PMD_INIT_LOG(ERR, "At most %d intr queues supported",
+ VMXNET3_MAX_RX_QUEUES);
+ return -ENOTSUP;
+ }
+
+ if (rte_intr_efd_enable(intr_handle, intr_vector)) {
+ PMD_INIT_LOG(ERR, "Failed to enable fastpath event fd");
+ return -1;
+ }
+
+ if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
+ intr_handle->intr_vec =
+ rte_zmalloc("intr_vec",
+ dev->data->nb_rx_queues * sizeof(int), 0);
+ if (intr_handle->intr_vec == NULL) {
+ PMD_INIT_LOG(ERR, "Failed to allocate %d Rx queues intr_vec",
+ dev->data->nb_rx_queues);
+ rte_intr_efd_disable(intr_handle);
+ return -ENOMEM;
+ }
+ }
+
+ if (!rte_intr_allow_others(intr_handle) &&
+ dev->data->dev_conf.intr_conf.lsc != 0) {
+ PMD_INIT_LOG(ERR, "not enough intr vector to support both Rx interrupt and LSC");
+ rte_free(intr_handle->intr_vec);
+ intr_handle->intr_vec = NULL;
+ rte_intr_efd_disable(intr_handle);
+ return -1;
+ }
+
+ /* if we cannot allocate one MSI-X vector per queue, don't enable
+ * interrupt mode.
+ */
+ if (hw->intr.num_intrs != (intr_handle->nb_efd + 1)) {
+ PMD_INIT_LOG(ERR, "Device configured with %d Rx intr vectors, expecting %d",
+ hw->intr.num_intrs, intr_handle->nb_efd + 1);
+ rte_free(intr_handle->intr_vec);
+ intr_handle->intr_vec = NULL;
+ rte_intr_efd_disable(intr_handle);
+ return -1;
+ }
+
+ for (i = 0; i < dev->data->nb_rx_queues; i++)
+ intr_handle->intr_vec[i] = i + 1;
+
+ for (i = 0; i < hw->intr.num_intrs; i++)
+ hw->intr.mod_levels[i] = UPT1_IML_ADAPTIVE;
+
+ PMD_INIT_LOG(INFO, "intr type %u, mode %u, %u vectors allocated",
+ hw->intr.type, hw->intr.mask_mode, hw->intr.num_intrs);
+
+ return 0;
+}
+
static int
vmxnet3_dev_setup_memreg(struct rte_eth_dev *dev)
{
@@ -605,6 +756,7 @@ vmxnet3_setup_driver_shared(struct rte_eth_dev *dev)
{
struct rte_eth_conf port_conf = dev->data->dev_conf;
struct vmxnet3_hw *hw = dev->data->dev_private;
+ struct rte_intr_handle *intr_handle = dev->intr_handle;
uint32_t mtu = dev->data->mtu;
Vmxnet3_DriverShared *shared = hw->shared;
Vmxnet3_DSDevRead *devRead = &shared->devRead;
@@ -630,15 +782,6 @@ vmxnet3_setup_driver_shared(struct rte_eth_dev *dev)
devRead->misc.numTxQueues = hw->num_tx_queues;
devRead->misc.numRxQueues = hw->num_rx_queues;
- /*
- * Set number of interrupts to 1
- * PMD by default disables all the interrupts but this is MUST
- * to activate device. It needs at least one interrupt for
- * link events to handle
- */
- hw->num_intrs = devRead->intrConf.numIntrs = 1;
- devRead->intrConf.intrCtrl |= VMXNET3_IC_DISABLE_ALL;
-
for (i = 0; i < hw->num_tx_queues; i++) {
Vmxnet3_TxQueueDesc *tqd = &hw->tqd_start[i];
vmxnet3_tx_queue_t *txq = dev->data->tx_queues[i];
@@ -655,9 +798,13 @@ vmxnet3_setup_driver_shared(struct rte_eth_dev *dev)
tqd->conf.compRingSize = txq->comp_ring.size;
tqd->conf.dataRingSize = txq->data_ring.size;
tqd->conf.txDataRingDescSize = txq->txdata_desc_size;
- tqd->conf.intrIdx = txq->comp_ring.intr_idx;
- tqd->status.stopped = TRUE;
- tqd->status.error = 0;
+
+ if (hw->intr.lsc_only)
+ tqd->conf.intrIdx = 1;
+ else
+ tqd->conf.intrIdx = intr_handle->intr_vec[i];
+ tqd->status.stopped = TRUE;
+ tqd->status.error = 0;
memset(&tqd->stats, 0, sizeof(tqd->stats));
}
@@ -674,16 +821,25 @@ vmxnet3_setup_driver_shared(struct rte_eth_dev *dev)
rqd->conf.rxRingSize[0] = rxq->cmd_ring[0].size;
rqd->conf.rxRingSize[1] = rxq->cmd_ring[1].size;
rqd->conf.compRingSize = rxq->comp_ring.size;
- rqd->conf.intrIdx = rxq->comp_ring.intr_idx;
- if (VMXNET3_VERSION_GE_3(hw)) {
- rqd->conf.rxDataRingBasePA = rxq->data_ring.basePA;
- rqd->conf.rxDataRingDescSize = rxq->data_desc_size;
- }
- rqd->status.stopped = TRUE;
- rqd->status.error = 0;
+
+ if (hw->intr.lsc_only)
+ rqd->conf.intrIdx = 1;
+ else
+ rqd->conf.intrIdx = intr_handle->intr_vec[i];
+ rqd->status.stopped = TRUE;
+ rqd->status.error = 0;
memset(&rqd->stats, 0, sizeof(rqd->stats));
}
+ /* intr settings */
+ devRead->intrConf.autoMask = hw->intr.mask_mode == VMXNET3_IMM_AUTO;
+ devRead->intrConf.numIntrs = hw->intr.num_intrs;
+ for (i = 0; i < hw->intr.num_intrs; i++)
+ devRead->intrConf.modLevels[i] = hw->intr.mod_levels[i];
+
+ devRead->intrConf.eventIntrIdx = hw->intr.event_intr_idx;
+ devRead->intrConf.intrCtrl |= rte_cpu_to_le_32(VMXNET3_IC_DISABLE_ALL);
+
/* RxMode set to 0 of VMXNET3_RXM_xxx */
devRead->rxFilterConf.rxMode = 0;
@@ -733,24 +889,18 @@ vmxnet3_dev_start(struct rte_eth_dev *dev)
/* Save stats before it is reset by CMD_ACTIVATE */
vmxnet3_hw_stats_save(hw);
+ /* configure MSI-X */
+ ret = vmxnet3_configure_msix(dev);
+ if (ret < 0) {
+ /* revert to lsc only */
+ hw->intr.num_intrs = 2;
+ hw->intr.lsc_only = TRUE;
+ }
+
ret = vmxnet3_setup_driver_shared(dev);
if (ret != VMXNET3_SUCCESS)
return ret;
- /* check if lsc interrupt feature is enabled */
- if (dev->data->dev_conf.intr_conf.lsc) {
- struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
-
- /* Setup interrupt callback */
- rte_intr_callback_register(&pci_dev->intr_handle,
- vmxnet3_interrupt_handler, dev);
-
- if (rte_intr_enable(&pci_dev->intr_handle) < 0) {
- PMD_INIT_LOG(ERR, "interrupt enable failed");
- return -EIO;
- }
- }
-
/* Exchange shared data with device */
VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAL,
VMXNET3_GET_ADDR_LO(hw->sharedPA));
@@ -790,9 +940,6 @@ vmxnet3_dev_start(struct rte_eth_dev *dev)
}
}
- /* Disable interrupts */
- vmxnet3_disable_intr(hw);
-
/*
* Load RX queues with blank mbufs and update next2fill index for device
* Update RxMode of the device
@@ -808,19 +955,29 @@ vmxnet3_dev_start(struct rte_eth_dev *dev)
/* Setting proper Rx Mode and issue Rx Mode Update command */
vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_UCAST | VMXNET3_RXM_BCAST, 1);
- if (dev->data->dev_conf.intr_conf.lsc) {
- vmxnet3_enable_intr(hw);
+ /* Setup interrupt callback */
+ rte_intr_callback_register(dev->intr_handle,
+ vmxnet3_interrupt_handler, dev);
- /*
- * Update link state from device since this won't be
- * done upon starting with lsc in use. This is done
- * only after enabling interrupts to avoid any race
- * where the link state could change without an
- * interrupt being fired.
- */
- __vmxnet3_dev_link_update(dev, 0);
+ if (rte_intr_enable(dev->intr_handle) < 0) {
+ PMD_INIT_LOG(ERR, "interrupt enable failed");
+ return -EIO;
}
+ /* enable all intrs */
+ vmxnet3_enable_all_intrs(hw);
+
+ vmxnet3_process_events(dev);
+
+ /*
+ * Update link state from device since this won't be
+ * done upon starting with lsc in use. This is done
+ * only after enabling interrupts to avoid any race
+ * where the link state could change without an
+ * interrupt being fired.
+ */
+ __vmxnet3_dev_link_update(dev, 0);
+
return VMXNET3_SUCCESS;
}
@@ -832,6 +989,8 @@ vmxnet3_dev_stop(struct rte_eth_dev *dev)
{
struct rte_eth_link link;
struct vmxnet3_hw *hw = dev->data->dev_private;
+ struct rte_intr_handle *intr_handle = dev->intr_handle;
+ int ret;
PMD_INIT_FUNC_TRACE();
@@ -840,16 +999,32 @@ vmxnet3_dev_stop(struct rte_eth_dev *dev)
return 0;
}
- /* disable interrupts */
- vmxnet3_disable_intr(hw);
+ do {
+ /* Unregister has lock to make sure there is no running cb.
+ * This has to happen first since vmxnet3_interrupt_handler
+ * reenables interrupts by calling vmxnet3_enable_intr
+ */
+ ret = rte_intr_callback_unregister(intr_handle,
+ vmxnet3_interrupt_handler,
+ (void *)-1);
+ } while (ret == -EAGAIN);
- if (dev->data->dev_conf.intr_conf.lsc) {
- struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
+ if (ret < 0)
+ PMD_DRV_LOG(ERR, "Error attempting to unregister intr cb: %d",
+ ret);
- rte_intr_disable(&pci_dev->intr_handle);
+ PMD_INIT_LOG(DEBUG, "Disabled %d intr callbacks", ret);
- rte_intr_callback_unregister(&pci_dev->intr_handle,
- vmxnet3_interrupt_handler, dev);
+ /* disable interrupts */
+ vmxnet3_disable_all_intrs(hw);
+
+ rte_intr_disable(intr_handle);
+
+ /* Clean datapath event and queue/vector mapping */
+ rte_intr_efd_disable(intr_handle);
+ if (intr_handle->intr_vec != NULL) {
+ rte_free(intr_handle->intr_vec);
+ intr_handle->intr_vec = NULL;
}
/* quiesce the device first */
@@ -1480,12 +1655,41 @@ static void
vmxnet3_interrupt_handler(void *param)
{
struct rte_eth_dev *dev = param;
- struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
+ struct vmxnet3_hw *hw = dev->data->dev_private;
+ Vmxnet3_DSDevRead *devRead = &hw->shared->devRead;
+ uint32_t events;
+
+ PMD_INIT_FUNC_TRACE();
+ vmxnet3_disable_intr(hw, devRead->intrConf.eventIntrIdx);
+
+ events = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_ECR);
+ if (events == 0)
+ goto done;
+ RTE_LOG(DEBUG, PMD, "Reading events: 0x%X", events);
vmxnet3_process_events(dev);
+done:
+ vmxnet3_enable_intr(hw, devRead->intrConf.eventIntrIdx);
+}
+
+static int
+vmxnet3_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
+{
+ struct vmxnet3_hw *hw = dev->data->dev_private;
+
+ vmxnet3_enable_intr(hw, dev->intr_handle->intr_vec[queue_id]);
- if (rte_intr_ack(&pci_dev->intr_handle) < 0)
- PMD_DRV_LOG(ERR, "interrupt enable failed");
+ return 0;
+}
+
+static int
+vmxnet3_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
+{
+ struct vmxnet3_hw *hw = dev->data->dev_private;
+
+ vmxnet3_disable_intr(hw, dev->intr_handle->intr_vec[queue_id]);
+
+ return 0;
}
RTE_PMD_REGISTER_PCI(net_vmxnet3, rte_vmxnet3_pmd);
diff --git a/drivers/net/vmxnet3/vmxnet3_ethdev.h b/drivers/net/vmxnet3/vmxnet3_ethdev.h
index f93bb474b2..59bee9723c 100644
--- a/drivers/net/vmxnet3/vmxnet3_ethdev.h
+++ b/drivers/net/vmxnet3/vmxnet3_ethdev.h
@@ -28,6 +28,8 @@
#define VMXNET3_RSS_MAX_KEY_SIZE 40
#define VMXNET3_RSS_MAX_IND_TABLE_SIZE 128
+#define VMXNET3_MAX_MSIX_VECT (VMXNET3_MAX_TX_QUEUES + \
+ VMXNET3_MAX_RX_QUEUES + 1)
#define VMXNET3_RSS_OFFLOAD_ALL ( \
ETH_RSS_IPV4 | \
@@ -63,6 +65,15 @@ typedef struct vmxnet3_mf_table {
uint16_t num_addrs; /* number of multicast addrs */
} vmxnet3_mf_table_t;
+struct vmxnet3_intr {
+ enum vmxnet3_intr_mask_mode mask_mode;
+ enum vmxnet3_intr_type type; /* MSI-X, MSI, or INTx? */
+ uint8_t num_intrs; /* # of intr vectors */
+ uint8_t event_intr_idx; /* idx of the intr vector for event */
+ uint8_t mod_levels[VMXNET3_MAX_MSIX_VECT]; /* moderation level */
+ bool lsc_only; /* no Rx queue interrupt */
+};
+
struct vmxnet3_hw {
uint8_t *hw_addr0; /* BAR0: PT-Passthrough Regs */
uint8_t *hw_addr1; /* BAR1: VD-Virtual Device Regs */
@@ -102,6 +113,7 @@ struct vmxnet3_hw {
uint64_t rss_confPA;
vmxnet3_mf_table_t *mf_table;
uint32_t shadow_vfta[VMXNET3_VFT_SIZE];
+ struct vmxnet3_intr intr;
Vmxnet3_MemRegs *memRegs;
uint64_t memRegsPA;
#define VMXNET3_VFT_TABLE_SIZE (VMXNET3_VFT_SIZE * sizeof(uint32_t))
--
2.17.1
More information about the dev
mailing list