Merge branch 'fix-race-conditions-in-ndo_get_stats64'

Shinas Rasheed says:

====================
Fix race conditions in ndo_get_stats64

Fix race conditions in ndo_get_stats64 by storing tx/rx stats
locally and not availing per queue resources which could be torn
down during interface stop. Also remove stats fetch from
firmware which is currently unnecessary
====================

Link: https://patch.msgid.link/20250117094653.2588578-1-srasheed@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This commit is contained in:
Jakub Kicinski 2025-01-18 17:26:42 -08:00
commit ec81a13d88
14 changed files with 79 additions and 101 deletions

View File

@ -150,17 +150,14 @@ octep_get_ethtool_stats(struct net_device *netdev,
iface_rx_stats,
iface_tx_stats);
for (q = 0; q < oct->num_oqs; q++) {
struct octep_iq *iq = oct->iq[q];
struct octep_oq *oq = oct->oq[q];
for (q = 0; q < OCTEP_MAX_QUEUES; q++) {
tx_packets += oct->stats_iq[q].instr_completed;
tx_bytes += oct->stats_iq[q].bytes_sent;
tx_busy_errors += oct->stats_iq[q].tx_busy;
tx_packets += iq->stats.instr_completed;
tx_bytes += iq->stats.bytes_sent;
tx_busy_errors += iq->stats.tx_busy;
rx_packets += oq->stats.packets;
rx_bytes += oq->stats.bytes;
rx_alloc_errors += oq->stats.alloc_failures;
rx_packets += oct->stats_oq[q].packets;
rx_bytes += oct->stats_oq[q].bytes;
rx_alloc_errors += oct->stats_oq[q].alloc_failures;
}
i = 0;
data[i++] = rx_packets;
@ -198,22 +195,18 @@ octep_get_ethtool_stats(struct net_device *netdev,
data[i++] = iface_rx_stats->err_pkts;
/* Per Tx Queue stats */
for (q = 0; q < oct->num_iqs; q++) {
struct octep_iq *iq = oct->iq[q];
data[i++] = iq->stats.instr_posted;
data[i++] = iq->stats.instr_completed;
data[i++] = iq->stats.bytes_sent;
data[i++] = iq->stats.tx_busy;
for (q = 0; q < OCTEP_MAX_QUEUES; q++) {
data[i++] = oct->stats_iq[q].instr_posted;
data[i++] = oct->stats_iq[q].instr_completed;
data[i++] = oct->stats_iq[q].bytes_sent;
data[i++] = oct->stats_iq[q].tx_busy;
}
/* Per Rx Queue stats */
for (q = 0; q < oct->num_oqs; q++) {
struct octep_oq *oq = oct->oq[q];
data[i++] = oq->stats.packets;
data[i++] = oq->stats.bytes;
data[i++] = oq->stats.alloc_failures;
for (q = 0; q < OCTEP_MAX_QUEUES; q++) {
data[i++] = oct->stats_oq[q].packets;
data[i++] = oct->stats_oq[q].bytes;
data[i++] = oct->stats_oq[q].alloc_failures;
}
}

View File

@ -822,7 +822,7 @@ static inline int octep_iq_full_check(struct octep_iq *iq)
if (unlikely(IQ_INSTR_SPACE(iq) >
OCTEP_WAKE_QUEUE_THRESHOLD)) {
netif_start_subqueue(iq->netdev, iq->q_no);
iq->stats.restart_cnt++;
iq->stats->restart_cnt++;
return 0;
}
@ -960,7 +960,7 @@ static netdev_tx_t octep_start_xmit(struct sk_buff *skb,
wmb();
/* Ring Doorbell to notify the NIC of new packets */
writel(iq->fill_cnt, iq->doorbell_reg);
iq->stats.instr_posted += iq->fill_cnt;
iq->stats->instr_posted += iq->fill_cnt;
iq->fill_cnt = 0;
return NETDEV_TX_OK;
@ -991,37 +991,24 @@ static netdev_tx_t octep_start_xmit(struct sk_buff *skb,
static void octep_get_stats64(struct net_device *netdev,
struct rtnl_link_stats64 *stats)
{
u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
struct octep_device *oct = netdev_priv(netdev);
u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
int q;
if (netif_running(netdev))
octep_ctrl_net_get_if_stats(oct,
OCTEP_CTRL_NET_INVALID_VFID,
&oct->iface_rx_stats,
&oct->iface_tx_stats);
tx_packets = 0;
tx_bytes = 0;
rx_packets = 0;
rx_bytes = 0;
for (q = 0; q < oct->num_oqs; q++) {
struct octep_iq *iq = oct->iq[q];
struct octep_oq *oq = oct->oq[q];
tx_packets += iq->stats.instr_completed;
tx_bytes += iq->stats.bytes_sent;
rx_packets += oq->stats.packets;
rx_bytes += oq->stats.bytes;
for (q = 0; q < OCTEP_MAX_QUEUES; q++) {
tx_packets += oct->stats_iq[q].instr_completed;
tx_bytes += oct->stats_iq[q].bytes_sent;
rx_packets += oct->stats_oq[q].packets;
rx_bytes += oct->stats_oq[q].bytes;
}
stats->tx_packets = tx_packets;
stats->tx_bytes = tx_bytes;
stats->rx_packets = rx_packets;
stats->rx_bytes = rx_bytes;
stats->multicast = oct->iface_rx_stats.mcast_pkts;
stats->rx_errors = oct->iface_rx_stats.err_pkts;
stats->collisions = oct->iface_tx_stats.xscol;
stats->tx_fifo_errors = oct->iface_tx_stats.undflw;
}
/**

View File

@ -257,11 +257,17 @@ struct octep_device {
/* Pointers to Octeon Tx queues */
struct octep_iq *iq[OCTEP_MAX_IQ];
/* Per iq stats */
struct octep_iq_stats stats_iq[OCTEP_MAX_IQ];
/* Rx queues (OQ: Output Queue) */
u16 num_oqs;
/* Pointers to Octeon Rx queues */
struct octep_oq *oq[OCTEP_MAX_OQ];
/* Per oq stats */
struct octep_oq_stats stats_oq[OCTEP_MAX_OQ];
/* Hardware port number of the PCIe interface */
u16 pcie_port;

View File

@ -87,7 +87,7 @@ static int octep_oq_refill(struct octep_device *oct, struct octep_oq *oq)
page = dev_alloc_page();
if (unlikely(!page)) {
dev_err(oq->dev, "refill: rx buffer alloc failed\n");
oq->stats.alloc_failures++;
oq->stats->alloc_failures++;
break;
}
@ -98,7 +98,7 @@ static int octep_oq_refill(struct octep_device *oct, struct octep_oq *oq)
"OQ-%d buffer refill: DMA mapping error!\n",
oq->q_no);
put_page(page);
oq->stats.alloc_failures++;
oq->stats->alloc_failures++;
break;
}
oq->buff_info[refill_idx].page = page;
@ -134,6 +134,7 @@ static int octep_setup_oq(struct octep_device *oct, int q_no)
oq->netdev = oct->netdev;
oq->dev = &oct->pdev->dev;
oq->q_no = q_no;
oq->stats = &oct->stats_oq[q_no];
oq->max_count = CFG_GET_OQ_NUM_DESC(oct->conf);
oq->ring_size_mask = oq->max_count - 1;
oq->buffer_size = CFG_GET_OQ_BUF_SIZE(oct->conf);
@ -443,7 +444,7 @@ static int __octep_oq_process_rx(struct octep_device *oct,
if (!skb) {
octep_oq_drop_rx(oq, buff_info,
&read_idx, &desc_used);
oq->stats.alloc_failures++;
oq->stats->alloc_failures++;
continue;
}
skb_reserve(skb, data_offset);
@ -494,8 +495,8 @@ static int __octep_oq_process_rx(struct octep_device *oct,
oq->host_read_idx = read_idx;
oq->refill_count += desc_used;
oq->stats.packets += pkt;
oq->stats.bytes += rx_bytes;
oq->stats->packets += pkt;
oq->stats->bytes += rx_bytes;
return pkt;
}

View File

@ -186,8 +186,8 @@ struct octep_oq {
*/
u8 __iomem *pkts_sent_reg;
/* Statistics for this OQ. */
struct octep_oq_stats stats;
/* Pointer to statistics for this OQ. */
struct octep_oq_stats *stats;
/* Packets pending to be processed */
u32 pkts_pending;

View File

@ -81,9 +81,9 @@ int octep_iq_process_completions(struct octep_iq *iq, u16 budget)
}
iq->pkts_processed += compl_pkts;
iq->stats.instr_completed += compl_pkts;
iq->stats.bytes_sent += compl_bytes;
iq->stats.sgentry_sent += compl_sg;
iq->stats->instr_completed += compl_pkts;
iq->stats->bytes_sent += compl_bytes;
iq->stats->sgentry_sent += compl_sg;
iq->flush_index = fi;
netdev_tx_completed_queue(iq->netdev_q, compl_pkts, compl_bytes);
@ -187,6 +187,7 @@ static int octep_setup_iq(struct octep_device *oct, int q_no)
iq->netdev = oct->netdev;
iq->dev = &oct->pdev->dev;
iq->q_no = q_no;
iq->stats = &oct->stats_iq[q_no];
iq->max_count = CFG_GET_IQ_NUM_DESC(oct->conf);
iq->ring_size_mask = iq->max_count - 1;
iq->fill_threshold = CFG_GET_IQ_DB_MIN(oct->conf);

View File

@ -170,8 +170,8 @@ struct octep_iq {
*/
u16 flush_index;
/* Statistics for this input queue. */
struct octep_iq_stats stats;
/* Pointer to statistics for this input queue. */
struct octep_iq_stats *stats;
/* Pointer to the Virtual Base addr of the input ring. */
struct octep_tx_desc_hw *desc_ring;

View File

@ -114,12 +114,9 @@ static void octep_vf_get_ethtool_stats(struct net_device *netdev,
iface_tx_stats = &oct->iface_tx_stats;
iface_rx_stats = &oct->iface_rx_stats;
for (q = 0; q < oct->num_oqs; q++) {
struct octep_vf_iq *iq = oct->iq[q];
struct octep_vf_oq *oq = oct->oq[q];
tx_busy_errors += iq->stats.tx_busy;
rx_alloc_errors += oq->stats.alloc_failures;
for (q = 0; q < OCTEP_VF_MAX_QUEUES; q++) {
tx_busy_errors += oct->stats_iq[q].tx_busy;
rx_alloc_errors += oct->stats_oq[q].alloc_failures;
}
i = 0;
data[i++] = rx_alloc_errors;
@ -134,22 +131,18 @@ static void octep_vf_get_ethtool_stats(struct net_device *netdev,
data[i++] = iface_rx_stats->dropped_octets_fifo_full;
/* Per Tx Queue stats */
for (q = 0; q < oct->num_iqs; q++) {
struct octep_vf_iq *iq = oct->iq[q];
data[i++] = iq->stats.instr_posted;
data[i++] = iq->stats.instr_completed;
data[i++] = iq->stats.bytes_sent;
data[i++] = iq->stats.tx_busy;
for (q = 0; q < OCTEP_VF_MAX_QUEUES; q++) {
data[i++] = oct->stats_iq[q].instr_posted;
data[i++] = oct->stats_iq[q].instr_completed;
data[i++] = oct->stats_iq[q].bytes_sent;
data[i++] = oct->stats_iq[q].tx_busy;
}
/* Per Rx Queue stats */
for (q = 0; q < oct->num_oqs; q++) {
struct octep_vf_oq *oq = oct->oq[q];
data[i++] = oq->stats.packets;
data[i++] = oq->stats.bytes;
data[i++] = oq->stats.alloc_failures;
data[i++] = oct->stats_oq[q].packets;
data[i++] = oct->stats_oq[q].bytes;
data[i++] = oct->stats_oq[q].alloc_failures;
}
}

View File

@ -574,7 +574,7 @@ static int octep_vf_iq_full_check(struct octep_vf_iq *iq)
* caused queues to get re-enabled after
* being stopped
*/
iq->stats.restart_cnt++;
iq->stats->restart_cnt++;
fallthrough;
case 1: /* Queue left enabled, since IQ is not yet full*/
return 0;
@ -731,7 +731,7 @@ static netdev_tx_t octep_vf_start_xmit(struct sk_buff *skb,
/* Flush the hw descriptors before writing to doorbell */
smp_wmb();
writel(iq->fill_cnt, iq->doorbell_reg);
iq->stats.instr_posted += iq->fill_cnt;
iq->stats->instr_posted += iq->fill_cnt;
iq->fill_cnt = 0;
return NETDEV_TX_OK;
}
@ -786,27 +786,16 @@ static void octep_vf_get_stats64(struct net_device *netdev,
tx_bytes = 0;
rx_packets = 0;
rx_bytes = 0;
for (q = 0; q < oct->num_oqs; q++) {
struct octep_vf_iq *iq = oct->iq[q];
struct octep_vf_oq *oq = oct->oq[q];
tx_packets += iq->stats.instr_completed;
tx_bytes += iq->stats.bytes_sent;
rx_packets += oq->stats.packets;
rx_bytes += oq->stats.bytes;
for (q = 0; q < OCTEP_VF_MAX_QUEUES; q++) {
tx_packets += oct->stats_iq[q].instr_completed;
tx_bytes += oct->stats_iq[q].bytes_sent;
rx_packets += oct->stats_oq[q].packets;
rx_bytes += oct->stats_oq[q].bytes;
}
stats->tx_packets = tx_packets;
stats->tx_bytes = tx_bytes;
stats->rx_packets = rx_packets;
stats->rx_bytes = rx_bytes;
if (!octep_vf_get_if_stats(oct)) {
stats->multicast = oct->iface_rx_stats.mcast_pkts;
stats->rx_errors = oct->iface_rx_stats.err_pkts;
stats->rx_dropped = oct->iface_rx_stats.dropped_pkts_fifo_full +
oct->iface_rx_stats.err_pkts;
stats->rx_missed_errors = oct->iface_rx_stats.dropped_pkts_fifo_full;
stats->tx_dropped = oct->iface_tx_stats.dropped;
}
}
/**

View File

@ -246,11 +246,17 @@ struct octep_vf_device {
/* Pointers to Octeon Tx queues */
struct octep_vf_iq *iq[OCTEP_VF_MAX_IQ];
/* Per iq stats */
struct octep_vf_iq_stats stats_iq[OCTEP_VF_MAX_IQ];
/* Rx queues (OQ: Output Queue) */
u16 num_oqs;
/* Pointers to Octeon Rx queues */
struct octep_vf_oq *oq[OCTEP_VF_MAX_OQ];
/* Per oq stats */
struct octep_vf_oq_stats stats_oq[OCTEP_VF_MAX_OQ];
/* Hardware port number of the PCIe interface */
u16 pcie_port;

View File

@ -87,7 +87,7 @@ static int octep_vf_oq_refill(struct octep_vf_device *oct, struct octep_vf_oq *o
page = dev_alloc_page();
if (unlikely(!page)) {
dev_err(oq->dev, "refill: rx buffer alloc failed\n");
oq->stats.alloc_failures++;
oq->stats->alloc_failures++;
break;
}
@ -98,7 +98,7 @@ static int octep_vf_oq_refill(struct octep_vf_device *oct, struct octep_vf_oq *o
"OQ-%d buffer refill: DMA mapping error!\n",
oq->q_no);
put_page(page);
oq->stats.alloc_failures++;
oq->stats->alloc_failures++;
break;
}
oq->buff_info[refill_idx].page = page;
@ -134,6 +134,7 @@ static int octep_vf_setup_oq(struct octep_vf_device *oct, int q_no)
oq->netdev = oct->netdev;
oq->dev = &oct->pdev->dev;
oq->q_no = q_no;
oq->stats = &oct->stats_oq[q_no];
oq->max_count = CFG_GET_OQ_NUM_DESC(oct->conf);
oq->ring_size_mask = oq->max_count - 1;
oq->buffer_size = CFG_GET_OQ_BUF_SIZE(oct->conf);
@ -458,8 +459,8 @@ static int __octep_vf_oq_process_rx(struct octep_vf_device *oct,
oq->host_read_idx = read_idx;
oq->refill_count += desc_used;
oq->stats.packets += pkt;
oq->stats.bytes += rx_bytes;
oq->stats->packets += pkt;
oq->stats->bytes += rx_bytes;
return pkt;
}

View File

@ -187,7 +187,7 @@ struct octep_vf_oq {
u8 __iomem *pkts_sent_reg;
/* Statistics for this OQ. */
struct octep_vf_oq_stats stats;
struct octep_vf_oq_stats *stats;
/* Packets pending to be processed */
u32 pkts_pending;

View File

@ -82,9 +82,9 @@ int octep_vf_iq_process_completions(struct octep_vf_iq *iq, u16 budget)
}
iq->pkts_processed += compl_pkts;
iq->stats.instr_completed += compl_pkts;
iq->stats.bytes_sent += compl_bytes;
iq->stats.sgentry_sent += compl_sg;
iq->stats->instr_completed += compl_pkts;
iq->stats->bytes_sent += compl_bytes;
iq->stats->sgentry_sent += compl_sg;
iq->flush_index = fi;
netif_subqueue_completed_wake(iq->netdev, iq->q_no, compl_pkts,
@ -186,6 +186,7 @@ static int octep_vf_setup_iq(struct octep_vf_device *oct, int q_no)
iq->netdev = oct->netdev;
iq->dev = &oct->pdev->dev;
iq->q_no = q_no;
iq->stats = &oct->stats_iq[q_no];
iq->max_count = CFG_GET_IQ_NUM_DESC(oct->conf);
iq->ring_size_mask = iq->max_count - 1;
iq->fill_threshold = CFG_GET_IQ_DB_MIN(oct->conf);

View File

@ -129,7 +129,7 @@ struct octep_vf_iq {
u16 flush_index;
/* Statistics for this input queue. */
struct octep_vf_iq_stats stats;
struct octep_vf_iq_stats *stats;
/* Pointer to the Virtual Base addr of the input ring. */
struct octep_vf_tx_desc_hw *desc_ring;