Re: [dpdk-dev] [v8] net/e1000: fix i219 hang on reset/close

["Zhang, Xiao" <xiao.zhang@intel.com>]

Hi Gavin,

> -----Original Message-----
> From: Gavin Hu (Arm Technology China) [mailto:Gavin.Hu@arm.com]
> Sent: Thursday, September 5, 2019 2:07 PM
> To: Zhang, Xiao <xiao.zhang@intel.com>; dev@dpdk.org
> Cc: Lu, Wenzhuo <wenzhuo.lu@intel.com>; Zhao1, Wei <wei.zhao1@intel.com>;
> Ye, Xiaolong <xiaolong.ye@intel.com>; stable@dpdk.org
> Subject: RE: [dpdk-dev] [v8] net/e1000: fix i219 hang on reset/close
> 
> Hi Xiao,
> 
> > -----Original Message-----
> > From: dev <dev-bounces@dpdk.org> On Behalf Of Xiao Zhang
> > Sent: Monday, July 22, 2019 11:12 PM
> > To: dev@dpdk.org
> > Cc: wenzhuo.lu@intel.com; wei.zhao1@intel.com; xiaolong.ye@intel.com;
> > Xiao Zhang <xiao.zhang@intel.com>; stable@dpdk.org
> > Subject: [dpdk-dev] [v8] net/e1000: fix i219 hang on reset/close
> >
> > Unit hang may occur if multiple descriptors are available in the rings
> > during reset or close. This state can be detected by configure status
> > by bit 8 in register. If the bit is set and there are pending
> > descriptors in one of the rings, we must flush them before reset or
> > close.
> >
> > Fixes: 80580344("e1000: support EM devices (also known as
> > e1000/e1000e)")
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Xiao Zhang <xiao.zhang@intel.com>
> > ---
> > v8 Modify to follow code style of dpdk community.
> > v7 Add fix line.
> > v6 Change the fix on em driver instead of igb driver and update the
> > register address according to C-Spec.
> > v5 Change the subject.
> > v4 Correct the tail descriptor of tx ring.
> > v3 Add loop to handle all tx and rx queues.
> > v2 Use configuration register instead of NVM7 to get the hang state.
> > ---
> >  drivers/net/e1000/e1000_ethdev.h |   4 ++
> >  drivers/net/e1000/em_ethdev.c    |   5 ++
> >  drivers/net/e1000/em_rxtx.c      | 111
> > +++++++++++++++++++++++++++++++++++++++
> >  3 files changed, 120 insertions(+)
> >
> > diff --git a/drivers/net/e1000/e1000_ethdev.h
> > b/drivers/net/e1000/e1000_ethdev.h
> > index 67acb73..01ff943 100644
> > --- a/drivers/net/e1000/e1000_ethdev.h
> > +++ b/drivers/net/e1000/e1000_ethdev.h
> > @@ -35,6 +35,9 @@
> >  #define IGB_MAX_RX_QUEUE_NUM           8
> >  #define IGB_MAX_RX_QUEUE_NUM_82576     16
> >
> > +#define E1000_I219_MAX_RX_QUEUE_NUM          2
> > +#define E1000_I219_MAX_TX_QUEUE_NUM          2
> > +
> >  #define E1000_SYN_FILTER_ENABLE        0x00000001 /* syn filter enable field
> > */
> >  #define E1000_SYN_FILTER_QUEUE         0x0000000E /* syn filter queue field
> > */
> >  #define E1000_SYN_FILTER_QUEUE_SHIFT   1          /* syn filter queue field */
> > @@ -522,5 +525,6 @@ int igb_action_rss_same(const struct
> > rte_flow_action_rss *comp,  int igb_config_rss_filter(struct
> > rte_eth_dev *dev,
> >                       struct igb_rte_flow_rss_conf *conf,
> >                       bool add);
> > +void em_flush_desc_rings(struct rte_eth_dev *dev);
> >
> >  #endif /* _E1000_ETHDEV_H_ */
> > diff --git a/drivers/net/e1000/em_ethdev.c
> > b/drivers/net/e1000/em_ethdev.c index dc88661..62d3a95 100644
> > --- a/drivers/net/e1000/em_ethdev.c
> > +++ b/drivers/net/e1000/em_ethdev.c
> > @@ -738,6 +738,11 @@ eth_em_stop(struct rte_eth_dev *dev)
> >       em_lsc_intr_disable(hw);
> >
> >       e1000_reset_hw(hw);
> > +
> > +     /* Flush desc rings for i219 */
> > +     if (hw->mac.type >= e1000_pch_spt)
> > +             em_flush_desc_rings(dev);
> > +
> >       if (hw->mac.type >= e1000_82544)
> >               E1000_WRITE_REG(hw, E1000_WUC, 0);
> >
> > diff --git a/drivers/net/e1000/em_rxtx.c b/drivers/net/e1000/em_rxtx.c
> > index 708f832..55d8a67 100644
> > --- a/drivers/net/e1000/em_rxtx.c
> > +++ b/drivers/net/e1000/em_rxtx.c
> > @@ -18,6 +18,7 @@
> >  #include <rte_log.h>
> >  #include <rte_debug.h>
> >  #include <rte_pci.h>
> > +#include <rte_bus_pci.h>
> >  #include <rte_memory.h>
> >  #include <rte_memcpy.h>
> >  #include <rte_memzone.h>
> > @@ -59,6 +60,11 @@
> >  #define E1000_TX_OFFLOAD_NOTSUP_MASK \
> >               (PKT_TX_OFFLOAD_MASK ^ E1000_TX_OFFLOAD_MASK)
> >
> > +/* PCI offset for querying configuration status register */
> > +#define PCI_CFG_STATUS_REG                 0x06
> > +#define FLUSH_DESC_REQUIRED               0x100
> > +
> > +
> >  /**
> >   * Structure associated with each descriptor of the RX ring of a RX queue.
> >   */
> > @@ -2000,3 +2006,108 @@ em_txq_info_get(struct rte_eth_dev *dev,
> > uint16_t queue_id,
> >       qinfo->conf.tx_rs_thresh = txq->tx_rs_thresh;
> >       qinfo->conf.offloads = txq->offloads;  }
> > +
> > +static void
> > +e1000_flush_tx_ring(struct rte_eth_dev *dev) {
> > +     struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data-
> > >dev_private);
> > +     volatile struct e1000_data_desc *tx_desc;
> > +     volatile uint32_t *tdt_reg_addr;
> > +     uint32_t tdt, tctl, txd_lower = E1000_TXD_CMD_IFCS;
> > +     uint16_t size = 512;
> > +     struct em_tx_queue *txq;
> > +     int i;
> > +
> > +     if (dev->data->tx_queues == NULL)
> > +             return;
> > +     tctl = E1000_READ_REG(hw, E1000_TCTL);
> > +     E1000_WRITE_REG(hw, E1000_TCTL, tctl | E1000_TCTL_EN);
> > +     for (i = 0; i < dev->data->nb_tx_queues &&
> > +             i < E1000_I219_MAX_TX_QUEUE_NUM; i++) {
> > +             txq = dev->data->tx_queues[i];
> > +             tdt = E1000_READ_REG(hw, E1000_TDT(i));
> > +             if (tdt != txq->tx_tail)
> > +                     return;
> > +             tx_desc = &txq->tx_ring[txq->tx_tail];
> > +             tx_desc->buffer_addr = rte_cpu_to_le_64(txq-
> > >tx_ring_phys_addr);
> > +             tx_desc->lower.data = rte_cpu_to_le_32(txd_lower | size);
> > +             tx_desc->upper.data = 0;
> > +
> > +             rte_wmb();
> Tx_desc is CIO memory, should rte_cio_wmb be used here?
> 

Yes, will replay with rte_cio_wmb.

> > +             txq->tx_tail++;
> > +             if (txq->tx_tail == txq->nb_tx_desc)
> > +                     txq->tx_tail = 0;
> > +             rte_io_wmb();
> This line can be merged into the following line and use E1000_PCI_REG_WRITE
> API instead, which has an rte_io_wmb included.
> This not only looks better but also better for understanding.

Yes it's better to remove rte_io_wmb() and replace E1000_PCI_WRITE_RELAXED by
E1000_PCI_REG_WRITE.

> > +             tdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_TDT(i));
> > +             E1000_PCI_REG_WRITE_RELAXED(tdt_reg_addr, txq->tx_tail);
> > +             usec_delay(250);
> > +     }
> > +}
> > +
> > +static void
> > +e1000_flush_rx_ring(struct rte_eth_dev *dev) {
> > +     uint32_t rctl, rxdctl;
> > +     struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data-
> > >dev_private);
> > +     int i;
> > +
> > +     rctl = E1000_READ_REG(hw, E1000_RCTL);
> > +     E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
> > +     E1000_WRITE_FLUSH(hw);
> > +     usec_delay(150);
> > +
> > +     for (i = 0; i < dev->data->nb_rx_queues &&
> > +             i < E1000_I219_MAX_RX_QUEUE_NUM; i++) {
> > +             rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i));
> > +             /* zero the lower 14 bits (prefetch and host thresholds) */
> > +             rxdctl &= 0xffffc000;
> > +
> > +             /* update thresholds: prefetch threshold to 31,
> > +              * host threshold to 1 and make sure the granularity
> > +              * is "descriptors" and not "cache lines"
> > +              */
> > +             rxdctl |= (0x1F | (1UL << 8) |
> > E1000_RXDCTL_THRESH_UNIT_DESC);
> > +
> > +             E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl);
> > +     }
> > +     /* momentarily enable the RX ring for the changes to take effect */
> > +     E1000_WRITE_REG(hw, E1000_RCTL, rctl | E1000_RCTL_EN);
> > +     E1000_WRITE_FLUSH(hw);
> > +     usec_delay(150);
> > +     E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN); }
> > +
> > +/**
> > + * em_flush_desc_rings - remove all descriptors from the descriptor
> > +rings
> > + *
> > + * In i219, the descriptor rings must be emptied before
> > +resetting/closing the
> > + * HW. Failure to do this will cause the HW to enter a unit hang
> > +state which
> > + * can only be released by PCI reset on the device
> > + *
> > + */
> > +
> > +void
> > +em_flush_desc_rings(struct rte_eth_dev *dev) {
> > +     uint32_t fextnvm11, tdlen;
> > +     struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data-
> > >dev_private);
> > +     struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
> > +     uint16_t pci_cfg_status = 0;
> > +
> > +     fextnvm11 = E1000_READ_REG(hw, E1000_FEXTNVM11);
> > +     E1000_WRITE_REG(hw, E1000_FEXTNVM11,
> > +                     fextnvm11 |
> > E1000_FEXTNVM11_DISABLE_MULR_FIX);
> > +     tdlen = E1000_READ_REG(hw, E1000_TDLEN(0));
> > +     rte_pci_read_config(pci_dev, &pci_cfg_status, sizeof(pci_cfg_status),
> > +                             PCI_CFG_STATUS_REG);
> > +
> > +     /* do nothing if we're not in faulty state, or if the queue is empty */
> > +     if ((pci_cfg_status & FLUSH_DESC_REQUIRED) && tdlen) {
> > +             /* flush desc ring */
> > +             e1000_flush_tx_ring(dev);
> > +             rte_pci_read_config(pci_dev, &pci_cfg_status,
> > +                             sizeof(pci_cfg_status),
> > PCI_CFG_STATUS_REG);
> > +             if (pci_cfg_status & FLUSH_DESC_REQUIRED)
> > +                     e1000_flush_rx_ring(dev);
> > +     }
> > +}
> > --
> > 2.7.4
> 
> IMPORTANT NOTICE: The contents of this email and any attachments are
> confidential and may also be privileged. If you are not the intended recipient,
> please notify the sender immediately and do not disclose the contents to any
> other person, use it for any purpose, or store or copy the information in any
> medium. Thank you.