DPDK patches and discussions
 help / color / mirror / Atom feed
From: Yong Wang <yongwang@vmware.com>
To: "sahithi.singam@oracle.com" <sahithi.singam@oracle.com>
Cc: "dev@dpdk.org" <dev@dpdk.org>
Subject: Re: [PATCH] net/vmxnet3: add spinlocks to register command access
Date: Tue, 30 Nov 2021 07:31:31 +0000	[thread overview]
Message-ID: <EAB9C439-A6FD-415E-97BC-116A930CA634@vmware.com> (raw)
In-Reply-To: <20211108082331.1407-1-sahithi.singam@oracle.com>

-----Original Message-----
From: "sahithi.singam@oracle.com" <sahithi.singam@oracle.com>
Date: Monday, November 8, 2021 at 12:23 AM
To: Yong Wang <yongwang@vmware.com>
Cc: "dev@dpdk.org" <dev@dpdk.org>, Sahithi Singam <sahithi.singam@oracle.com>
Subject: [PATCH] net/vmxnet3: add spinlocks to register command access

    From: Sahithi Singam <sahithi.singam@oracle.com>

    At present, there are no spinlocks around register command access.
    This resulted in a race condition when two threads running on
    two different cores invoked link_update function at the same time
    to get link status. Due to this race condition, one of the threads
    reported false link status value.

    Signed-off-by: Sahithi Singam <sahithi.singam@oracle.com>
    ---

Thanks Sahithi for the patch.  As we discussed offline, in DPDK, the expectation is that control level synchronization should be handled by the application.  In my knowledge, currently no PMD guarantee such synchronization at driver callback level.  It makes more sense to have the application manages the synchronization as most likely it needs to work with multiple PMDs and it's better to keep this behavior consistent across all PMDs (i.e, it does not make a lot of sense to support this behavior only in one particular PMD).

     drivers/net/vmxnet3/vmxnet3_ethdev.c | 37 ++++++++++++++++++++++++++++
     drivers/net/vmxnet3/vmxnet3_ethdev.h |  1 +
     drivers/net/vmxnet3/vmxnet3_rxtx.c   |  2 ++
     3 files changed, 40 insertions(+)

    diff --git a/drivers/net/vmxnet3/vmxnet3_ethdev.c b/drivers/net/vmxnet3/vmxnet3_ethdev.c
    index d1ef1cad08..d4a433e0db 100644
    --- a/drivers/net/vmxnet3/vmxnet3_ethdev.c
    +++ b/drivers/net/vmxnet3/vmxnet3_ethdev.c
    @@ -252,9 +252,11 @@ eth_vmxnet3_txdata_get(struct vmxnet3_hw *hw)
     {
     	uint16 txdata_desc_size;

    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
     			       VMXNET3_CMD_GET_TXDATA_DESC_SIZE);
     	txdata_desc_size = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
    +	rte_spinlock_unlock(&hw->cmd_lock);

     	return (txdata_desc_size < VMXNET3_TXDATA_DESC_MIN_SIZE ||
     		txdata_desc_size > VMXNET3_TXDATA_DESC_MAX_SIZE ||
    @@ -285,6 +287,7 @@ eth_vmxnet3_dev_init(struct rte_eth_dev *eth_dev)
     	eth_dev->tx_pkt_burst = &vmxnet3_xmit_pkts;
     	eth_dev->tx_pkt_prepare = vmxnet3_prep_pkts;
     	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
    +	rte_spinlock_init(&hw->cmd_lock);

     	/* extra mbuf field is required to guess MSS */
     	vmxnet3_segs_dynfield_offset =
    @@ -375,7 +378,9 @@ eth_vmxnet3_dev_init(struct rte_eth_dev *eth_dev)
     		     hw->perm_addr[3], hw->perm_addr[4], hw->perm_addr[5]);

     	/* Put device in Quiesce Mode */
    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_QUIESCE_DEV);
    +	rte_spinlock_unlock(&hw->cmd_lock);

     	/* allow untagged pkts */
     	VMXNET3_SET_VFTABLE_ENTRY(hw->shadow_vfta, 0);
    @@ -451,9 +456,11 @@ vmxnet3_alloc_intr_resources(struct rte_eth_dev *dev)
     	int nvec = 1; /* for link event */

     	/* intr settings */
    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
     			       VMXNET3_CMD_GET_CONF_INTR);
     	cfg = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
    +	rte_spinlock_unlock(&hw->cmd_lock);
     	hw->intr.type = cfg & 0x3;
     	hw->intr.mask_mode = (cfg >> 2) & 0x3;

    @@ -910,8 +917,10 @@ vmxnet3_dev_start(struct rte_eth_dev *dev)
     			       VMXNET3_GET_ADDR_HI(hw->sharedPA));

     	/* Activate device by register write */
    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_ACTIVATE_DEV);
     	ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
    +	rte_spinlock_unlock(&hw->cmd_lock);

     	if (ret != 0) {
     		PMD_INIT_LOG(ERR, "Device activation: UNSUCCESSFUL");
    @@ -921,9 +930,11 @@ vmxnet3_dev_start(struct rte_eth_dev *dev)
     	/* Setup memory region for rx buffers */
     	ret = vmxnet3_dev_setup_memreg(dev);
     	if (ret == 0) {
    +		rte_spinlock_lock(&hw->cmd_lock);
     		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
     				       VMXNET3_CMD_REGISTER_MEMREGS);
     		ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
    +		rte_spinlock_unlock(&hw->cmd_lock);
     		if (ret != 0)
     			PMD_INIT_LOG(DEBUG,
     				     "Failed in setup memory region cmd\n");
    @@ -1027,12 +1038,16 @@ vmxnet3_dev_stop(struct rte_eth_dev *dev)
     	rte_intr_vec_list_free(intr_handle);

     	/* quiesce the device first */
    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_QUIESCE_DEV);
    +	rte_spinlock_unlock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAL, 0);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAH, 0);

     	/* reset the device */
    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
    +	rte_spinlock_unlock(&hw->cmd_lock);
     	PMD_INIT_LOG(DEBUG, "Device reset.");

     	vmxnet3_dev_clear_queues(dev);
    @@ -1182,7 +1197,9 @@ vmxnet3_hw_stats_save(struct vmxnet3_hw *hw)
     {
     	unsigned int i;

    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_STATS);
    +	rte_spinlock_unlock(&hw->cmd_lock);

     	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_TX_QUEUES);

    @@ -1285,7 +1302,9 @@ vmxnet3_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
     	struct UPT1_TxStats txStats;
     	struct UPT1_RxStats rxStats;

    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_STATS);
    +	rte_spinlock_unlock(&hw->cmd_lock);

     	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_TX_QUEUES);
     	for (i = 0; i < hw->num_tx_queues; i++) {
    @@ -1335,7 +1354,9 @@ vmxnet3_dev_stats_reset(struct rte_eth_dev *dev)
     	struct UPT1_TxStats txStats = {0};
     	struct UPT1_RxStats rxStats = {0};

    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_STATS);
    +	rte_spinlock_unlock(&hw->cmd_lock);

     	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_TX_QUEUES);

    @@ -1443,8 +1464,10 @@ __vmxnet3_dev_link_update(struct rte_eth_dev *dev,

     	memset(&link, 0, sizeof(link));

    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
     	ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
    +	rte_spinlock_unlock(&hw->cmd_lock);

     	if (ret & 0x1)
     		link.link_status = RTE_ETH_LINK_UP;
    @@ -1476,7 +1499,9 @@ vmxnet3_dev_set_rxmode(struct vmxnet3_hw *hw, uint32_t feature, int set)
     	else
     		rxConf->rxMode = rxConf->rxMode & (~feature);

    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_UPDATE_RX_MODE);
    +	rte_spinlock_unlock(&hw->cmd_lock);
     }

     /* Promiscuous supported only if Vmxnet3_DriverShared is initialized in adapter */
    @@ -1489,8 +1514,10 @@ vmxnet3_dev_promiscuous_enable(struct rte_eth_dev *dev)
     	memset(vf_table, 0, VMXNET3_VFT_TABLE_SIZE);
     	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_PROMISC, 1);

    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
     			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
    +	rte_spinlock_unlock(&hw->cmd_lock);

     	return 0;
     }
    @@ -1508,8 +1535,10 @@ vmxnet3_dev_promiscuous_disable(struct rte_eth_dev *dev)
     	else
     		memset(vf_table, 0xff, VMXNET3_VFT_TABLE_SIZE);
     	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_PROMISC, 0);
    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
     			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
    +	rte_spinlock_unlock(&hw->cmd_lock);

     	return 0;
     }
    @@ -1560,8 +1589,10 @@ vmxnet3_dev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vid, int on)
     	else
     		VMXNET3_CLEAR_VFTABLE_ENTRY(vf_table, vid);

    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
     			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
    +	rte_spinlock_unlock(&hw->cmd_lock);
     	return 0;
     }

    @@ -1579,8 +1610,10 @@ vmxnet3_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
     		else
     			devRead->misc.uptFeatures &= ~UPT1_F_RXVLAN;

    +		rte_spinlock_lock(&hw->cmd_lock);
     		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
     				       VMXNET3_CMD_UPDATE_FEATURE);
    +		rte_spinlock_unlock(&hw->cmd_lock);
     	}

     	if (mask & RTE_ETH_VLAN_FILTER_MASK) {
    @@ -1589,8 +1622,10 @@ vmxnet3_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
     		else
     			memset(vf_table, 0xff, VMXNET3_VFT_TABLE_SIZE);

    +		rte_spinlock_lock(&hw->cmd_lock);
     		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
     				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
    +		rte_spinlock_unlock(&hw->cmd_lock);
     	}

     	return 0;
    @@ -1622,8 +1657,10 @@ vmxnet3_process_events(struct rte_eth_dev *dev)

     	/* Check if there is an error on xmit/recv queues */
     	if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
    +		rte_spinlock_lock(&hw->cmd_lock);
     		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
     				       VMXNET3_CMD_GET_QUEUE_STATUS);
    +		rte_spinlock_unlock(&hw->cmd_lock);

     		if (hw->tqd_start->status.stopped)
     			PMD_DRV_LOG(ERR, "tq error 0x%x",
    diff --git a/drivers/net/vmxnet3/vmxnet3_ethdev.h b/drivers/net/vmxnet3/vmxnet3_ethdev.h
    index ef858ac951..d07a8f2757 100644
    --- a/drivers/net/vmxnet3/vmxnet3_ethdev.h
    +++ b/drivers/net/vmxnet3/vmxnet3_ethdev.h
    @@ -108,6 +108,7 @@ struct vmxnet3_hw {
     	uint64_t              queueDescPA;
     	uint16_t              queue_desc_len;
     	uint16_t              mtu;
    +	rte_spinlock_t        cmd_lock;

     	VMXNET3_RSSConf       *rss_conf;
     	uint64_t              rss_confPA;
    diff --git a/drivers/net/vmxnet3/vmxnet3_rxtx.c b/drivers/net/vmxnet3/vmxnet3_rxtx.c
    index deba64be6a..c87f1c6470 100644
    --- a/drivers/net/vmxnet3/vmxnet3_rxtx.c
    +++ b/drivers/net/vmxnet3/vmxnet3_rxtx.c
    @@ -1334,9 +1334,11 @@ vmxnet3_v4_rss_configure(struct rte_eth_dev *dev)
     	if (rss_hf & RTE_ETH_RSS_NONFRAG_IPV6_UDP)
     		cmdInfo->setRSSFields |= VMXNET3_RSS_FIELDS_UDPIP6;

    +	rte_spinlock_lock(&hw->cmd_lock);
     	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
     			       VMXNET3_CMD_SET_RSS_FIELDS);
     	ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
    +	rte_spinlock_unlock(&hw->cmd_lock);

     	if (ret != VMXNET3_SUCCESS) {
     		PMD_DRV_LOG(ERR, "Set RSS fields (v4) failed: %d", ret);
    -- 
    2.32.0.windows.1



  parent reply	other threads:[~2021-11-30  7:31 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-11-08  8:23 [dpdk-dev] " sahithi.singam
2021-11-16 17:16 ` Ferruh Yigit
2021-11-30  7:31 ` Yong Wang [this message]
2021-11-30  8:58   ` Ferruh Yigit

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=EAB9C439-A6FD-415E-97BC-116A930CA634@vmware.com \
    --to=yongwang@vmware.com \
    --cc=dev@dpdk.org \
    --cc=sahithi.singam@oracle.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).