From 04a1839950d92ab6519479bc95710e89ae6cbc77 Mon Sep 17 00:00:00 2001 From: Egor Pomozov Date: Tue, 22 Oct 2019 09:53:35 +0000 Subject: net: aquantia: implement data PTP datapath Here we do alloc/free IRQs for PTP rings. We also implement processing of PTP packets on TX and RX sides. Signed-off-by: Egor Pomozov Co-developed-by: Sergey Samoilenko Signed-off-by: Sergey Samoilenko Co-developed-by: Dmitry Bezrukov Signed-off-by: Dmitry Bezrukov Signed-off-by: Igor Russkikh Signed-off-by: David S. Miller --- drivers/net/ethernet/aquantia/atlantic/aq_ptp.c | 534 +++++++++++++++++++++++- 1 file changed, 532 insertions(+), 2 deletions(-) (limited to 'drivers/net/ethernet/aquantia/atlantic/aq_ptp.c') diff --git a/drivers/net/ethernet/aquantia/atlantic/aq_ptp.c b/drivers/net/ethernet/aquantia/atlantic/aq_ptp.c index f2fd0ca14a49..fbb1912a34d7 100644 --- a/drivers/net/ethernet/aquantia/atlantic/aq_ptp.c +++ b/drivers/net/ethernet/aquantia/atlantic/aq_ptp.c @@ -8,12 +8,24 @@ */ #include +#include #include #include "aq_nic.h" #include "aq_ptp.h" #include "aq_ring.h" +#define AQ_PTP_TX_TIMEOUT (HZ * 10) + +enum ptp_speed_offsets { + ptp_offset_idx_10 = 0, + ptp_offset_idx_100, + ptp_offset_idx_1000, + ptp_offset_idx_2500, + ptp_offset_idx_5000, + ptp_offset_idx_10000, +}; + struct ptp_skb_ring { struct sk_buff **buff; spinlock_t lock; @@ -22,6 +34,12 @@ struct ptp_skb_ring { unsigned int tail; }; +struct ptp_tx_timeout { + spinlock_t lock; + bool active; + unsigned long tx_start; +}; + struct aq_ptp_s { struct aq_nic_s *aq_nic; spinlock_t ptp_lock; @@ -29,8 +47,16 @@ struct aq_ptp_s { struct ptp_clock *ptp_clock; struct ptp_clock_info ptp_info; + atomic_t offset_egress; + atomic_t offset_ingress; + struct aq_ring_param_s ptp_ring_param; + struct ptp_tx_timeout ptp_tx_timeout; + + unsigned int idx_vector; + struct napi_struct napi; + struct aq_ring_s ptp_tx; struct aq_ring_s ptp_rx; struct aq_ring_s hwts_rx; @@ -38,6 +64,101 @@ struct aq_ptp_s { struct ptp_skb_ring skb_ring; }; +struct ptp_tm_offset { + unsigned int mbps; + int egress; + int ingress; +}; + +static struct ptp_tm_offset ptp_offset[6]; + +void aq_ptp_tm_offset_set(struct aq_nic_s *aq_nic, unsigned int mbps) +{ + struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp; + int i, egress, ingress; + + if (!aq_ptp) + return; + + egress = 0; + ingress = 0; + + for (i = 0; i < ARRAY_SIZE(ptp_offset); i++) { + if (mbps == ptp_offset[i].mbps) { + egress = ptp_offset[i].egress; + ingress = ptp_offset[i].ingress; + break; + } + } + + atomic_set(&aq_ptp->offset_egress, egress); + atomic_set(&aq_ptp->offset_ingress, ingress); +} + +static int __aq_ptp_skb_put(struct ptp_skb_ring *ring, struct sk_buff *skb) +{ + unsigned int next_head = (ring->head + 1) % ring->size; + + if (next_head == ring->tail) + return -ENOMEM; + + ring->buff[ring->head] = skb_get(skb); + ring->head = next_head; + + return 0; +} + +static int aq_ptp_skb_put(struct ptp_skb_ring *ring, struct sk_buff *skb) +{ + unsigned long flags; + int ret; + + spin_lock_irqsave(&ring->lock, flags); + ret = __aq_ptp_skb_put(ring, skb); + spin_unlock_irqrestore(&ring->lock, flags); + + return ret; +} + +static struct sk_buff *__aq_ptp_skb_get(struct ptp_skb_ring *ring) +{ + struct sk_buff *skb; + + if (ring->tail == ring->head) + return NULL; + + skb = ring->buff[ring->tail]; + ring->tail = (ring->tail + 1) % ring->size; + + return skb; +} + +static struct sk_buff *aq_ptp_skb_get(struct ptp_skb_ring *ring) +{ + unsigned long flags; + struct sk_buff *skb; + + spin_lock_irqsave(&ring->lock, flags); + skb = __aq_ptp_skb_get(ring); + spin_unlock_irqrestore(&ring->lock, flags); + + return skb; +} + +static unsigned int aq_ptp_skb_buf_len(struct ptp_skb_ring *ring) +{ + unsigned long flags; + unsigned int len; + + spin_lock_irqsave(&ring->lock, flags); + len = (ring->head >= ring->tail) ? + ring->head - ring->tail : + ring->size - ring->tail + ring->head; + spin_unlock_irqrestore(&ring->lock, flags); + + return len; +} + static int aq_ptp_skb_ring_init(struct ptp_skb_ring *ring, unsigned int size) { struct sk_buff **buff = kmalloc(sizeof(*buff) * size, GFP_KERNEL); @@ -55,10 +176,75 @@ static int aq_ptp_skb_ring_init(struct ptp_skb_ring *ring, unsigned int size) return 0; } +static void aq_ptp_skb_ring_clean(struct ptp_skb_ring *ring) +{ + struct sk_buff *skb; + + while ((skb = aq_ptp_skb_get(ring)) != NULL) + dev_kfree_skb_any(skb); +} + static void aq_ptp_skb_ring_release(struct ptp_skb_ring *ring) { - kfree(ring->buff); - ring->buff = NULL; + if (ring->buff) { + aq_ptp_skb_ring_clean(ring); + kfree(ring->buff); + ring->buff = NULL; + } +} + +static void aq_ptp_tx_timeout_init(struct ptp_tx_timeout *timeout) +{ + spin_lock_init(&timeout->lock); + timeout->active = false; +} + +static void aq_ptp_tx_timeout_start(struct aq_ptp_s *aq_ptp) +{ + struct ptp_tx_timeout *timeout = &aq_ptp->ptp_tx_timeout; + unsigned long flags; + + spin_lock_irqsave(&timeout->lock, flags); + timeout->active = true; + timeout->tx_start = jiffies; + spin_unlock_irqrestore(&timeout->lock, flags); +} + +static void aq_ptp_tx_timeout_update(struct aq_ptp_s *aq_ptp) +{ + if (!aq_ptp_skb_buf_len(&aq_ptp->skb_ring)) { + struct ptp_tx_timeout *timeout = &aq_ptp->ptp_tx_timeout; + unsigned long flags; + + spin_lock_irqsave(&timeout->lock, flags); + timeout->active = false; + spin_unlock_irqrestore(&timeout->lock, flags); + } +} + +static void aq_ptp_tx_timeout_check(struct aq_ptp_s *aq_ptp) +{ + struct ptp_tx_timeout *timeout = &aq_ptp->ptp_tx_timeout; + unsigned long flags; + bool timeout_flag; + + timeout_flag = false; + + spin_lock_irqsave(&timeout->lock, flags); + if (timeout->active) { + timeout_flag = time_is_before_jiffies(timeout->tx_start + + AQ_PTP_TX_TIMEOUT); + /* reset active flag if timeout detected */ + if (timeout_flag) + timeout->active = false; + } + spin_unlock_irqrestore(&timeout->lock, flags); + + if (timeout_flag) { + aq_ptp_skb_ring_clean(&aq_ptp->skb_ring); + netdev_err(aq_ptp->aq_nic->ndev, + "PTP Timeout. Clearing Tx Timestamp SKBs\n"); + } } /* aq_ptp_adjfine @@ -148,6 +334,263 @@ static int aq_ptp_settime(struct ptp_clock_info *ptp, return 0; } +static void aq_ptp_convert_to_hwtstamp(struct aq_ptp_s *aq_ptp, + struct skb_shared_hwtstamps *hwtstamp, + u64 timestamp) +{ + memset(hwtstamp, 0, sizeof(*hwtstamp)); + hwtstamp->hwtstamp = ns_to_ktime(timestamp); +} + +/* aq_ptp_tx_hwtstamp - utility function which checks for TX time stamp + * @adapter: the private adapter struct + * + * if the timestamp is valid, we convert it into the timecounter ns + * value, then store that result into the hwtstamps structure which + * is passed up the network stack + */ +void aq_ptp_tx_hwtstamp(struct aq_nic_s *aq_nic, u64 timestamp) +{ + struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp; + struct sk_buff *skb = aq_ptp_skb_get(&aq_ptp->skb_ring); + struct skb_shared_hwtstamps hwtstamp; + + if (!skb) { + netdev_err(aq_nic->ndev, "have timestamp but tx_queus empty\n"); + return; + } + + timestamp += atomic_read(&aq_ptp->offset_egress); + aq_ptp_convert_to_hwtstamp(aq_ptp, &hwtstamp, timestamp); + skb_tstamp_tx(skb, &hwtstamp); + dev_kfree_skb_any(skb); + + aq_ptp_tx_timeout_update(aq_ptp); +} + +/* aq_ptp_rx_hwtstamp - utility function which checks for RX time stamp + * @adapter: pointer to adapter struct + * @skb: particular skb to send timestamp with + * + * if the timestamp is valid, we convert it into the timecounter ns + * value, then store that result into the hwtstamps structure which + * is passed up the network stack + */ +static void aq_ptp_rx_hwtstamp(struct aq_ptp_s *aq_ptp, struct sk_buff *skb, + u64 timestamp) +{ + timestamp -= atomic_read(&aq_ptp->offset_ingress); + aq_ptp_convert_to_hwtstamp(aq_ptp, skb_hwtstamps(skb), timestamp); +} + +bool aq_ptp_ring(struct aq_nic_s *aq_nic, struct aq_ring_s *ring) +{ + struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp; + + if (!aq_ptp) + return false; + + return &aq_ptp->ptp_tx == ring || + &aq_ptp->ptp_rx == ring || &aq_ptp->hwts_rx == ring; +} + +u16 aq_ptp_extract_ts(struct aq_nic_s *aq_nic, struct sk_buff *skb, u8 *p, + unsigned int len) +{ + struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp; + u64 timestamp = 0; + u16 ret = aq_nic->aq_hw_ops->rx_extract_ts(aq_nic->aq_hw, + p, len, ×tamp); + + if (ret > 0) + aq_ptp_rx_hwtstamp(aq_ptp, skb, timestamp); + + return ret; +} + +static int aq_ptp_poll(struct napi_struct *napi, int budget) +{ + struct aq_ptp_s *aq_ptp = container_of(napi, struct aq_ptp_s, napi); + struct aq_nic_s *aq_nic = aq_ptp->aq_nic; + bool was_cleaned = false; + int work_done = 0; + int err; + + /* Processing PTP TX traffic */ + err = aq_nic->aq_hw_ops->hw_ring_tx_head_update(aq_nic->aq_hw, + &aq_ptp->ptp_tx); + if (err < 0) + goto err_exit; + + if (aq_ptp->ptp_tx.sw_head != aq_ptp->ptp_tx.hw_head) { + aq_ring_tx_clean(&aq_ptp->ptp_tx); + + was_cleaned = true; + } + + /* Processing HW_TIMESTAMP RX traffic */ + err = aq_nic->aq_hw_ops->hw_ring_hwts_rx_receive(aq_nic->aq_hw, + &aq_ptp->hwts_rx); + if (err < 0) + goto err_exit; + + if (aq_ptp->hwts_rx.sw_head != aq_ptp->hwts_rx.hw_head) { + aq_ring_hwts_rx_clean(&aq_ptp->hwts_rx, aq_nic); + + err = aq_nic->aq_hw_ops->hw_ring_hwts_rx_fill(aq_nic->aq_hw, + &aq_ptp->hwts_rx); + + was_cleaned = true; + } + + /* Processing PTP RX traffic */ + err = aq_nic->aq_hw_ops->hw_ring_rx_receive(aq_nic->aq_hw, + &aq_ptp->ptp_rx); + if (err < 0) + goto err_exit; + + if (aq_ptp->ptp_rx.sw_head != aq_ptp->ptp_rx.hw_head) { + unsigned int sw_tail_old; + + err = aq_ring_rx_clean(&aq_ptp->ptp_rx, napi, &work_done, budget); + if (err < 0) + goto err_exit; + + sw_tail_old = aq_ptp->ptp_rx.sw_tail; + err = aq_ring_rx_fill(&aq_ptp->ptp_rx); + if (err < 0) + goto err_exit; + + err = aq_nic->aq_hw_ops->hw_ring_rx_fill(aq_nic->aq_hw, + &aq_ptp->ptp_rx, + sw_tail_old); + if (err < 0) + goto err_exit; + } + + if (was_cleaned) + work_done = budget; + + if (work_done < budget) { + napi_complete_done(napi, work_done); + aq_nic->aq_hw_ops->hw_irq_enable(aq_nic->aq_hw, + 1 << aq_ptp->ptp_ring_param.vec_idx); + } + +err_exit: + return work_done; +} + +static irqreturn_t aq_ptp_isr(int irq, void *private) +{ + struct aq_ptp_s *aq_ptp = private; + int err = 0; + + if (!aq_ptp) { + err = -EINVAL; + goto err_exit; + } + napi_schedule(&aq_ptp->napi); + +err_exit: + return err >= 0 ? IRQ_HANDLED : IRQ_NONE; +} + +int aq_ptp_xmit(struct aq_nic_s *aq_nic, struct sk_buff *skb) +{ + struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp; + struct aq_ring_s *ring = &aq_ptp->ptp_tx; + unsigned long irq_flags; + int err = NETDEV_TX_OK; + unsigned int frags; + + if (skb->len <= 0) { + dev_kfree_skb_any(skb); + goto err_exit; + } + + frags = skb_shinfo(skb)->nr_frags + 1; + /* Frags cannot be bigger 16KB + * because PTP usually works + * without Jumbo even in a background + */ + if (frags > AQ_CFG_SKB_FRAGS_MAX || frags > aq_ring_avail_dx(ring)) { + /* Drop packet because it doesn't make sence to delay it */ + dev_kfree_skb_any(skb); + goto err_exit; + } + + err = aq_ptp_skb_put(&aq_ptp->skb_ring, skb); + if (err) { + netdev_err(aq_nic->ndev, "SKB Ring is overflow (%u)!\n", + ring->size); + return NETDEV_TX_BUSY; + } + skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; + aq_ptp_tx_timeout_start(aq_ptp); + skb_tx_timestamp(skb); + + spin_lock_irqsave(&aq_nic->aq_ptp->ptp_ring_lock, irq_flags); + frags = aq_nic_map_skb(aq_nic, skb, ring); + + if (likely(frags)) { + err = aq_nic->aq_hw_ops->hw_ring_tx_xmit(aq_nic->aq_hw, + ring, frags); + if (err >= 0) { + ++ring->stats.tx.packets; + ring->stats.tx.bytes += skb->len; + } + } else { + err = NETDEV_TX_BUSY; + } + spin_unlock_irqrestore(&aq_nic->aq_ptp->ptp_ring_lock, irq_flags); + +err_exit: + return err; +} + +void aq_ptp_service_task(struct aq_nic_s *aq_nic) +{ + struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp; + + if (!aq_ptp) + return; + + aq_ptp_tx_timeout_check(aq_ptp); +} + +int aq_ptp_irq_alloc(struct aq_nic_s *aq_nic) +{ + struct pci_dev *pdev = aq_nic->pdev; + struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp; + int err = 0; + + if (!aq_ptp) + return 0; + + if (pdev->msix_enabled || pdev->msi_enabled) { + err = request_irq(pci_irq_vector(pdev, aq_ptp->idx_vector), + aq_ptp_isr, 0, aq_nic->ndev->name, aq_ptp); + } else { + err = -EINVAL; + goto err_exit; + } + +err_exit: + return err; +} + +void aq_ptp_irq_free(struct aq_nic_s *aq_nic) +{ + struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp; + struct pci_dev *pdev = aq_nic->pdev; + + if (!aq_ptp) + return; + + free_irq(pci_irq_vector(pdev, aq_ptp->idx_vector), aq_ptp); +} + int aq_ptp_ring_init(struct aq_nic_s *aq_nic) { struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp; @@ -189,6 +632,12 @@ int aq_ptp_ring_init(struct aq_nic_s *aq_nic) err = aq_nic->aq_hw_ops->hw_ring_rx_init(aq_nic->aq_hw, &aq_ptp->hwts_rx, &aq_ptp->ptp_ring_param); + if (err < 0) + goto err_exit; + err = aq_nic->aq_hw_ops->hw_ring_hwts_rx_fill(aq_nic->aq_hw, + &aq_ptp->hwts_rx); + if (err < 0) + goto err_exit; return err; @@ -219,6 +668,8 @@ int aq_ptp_ring_start(struct aq_nic_s *aq_nic) if (err < 0) goto err_exit; + napi_enable(&aq_ptp->napi); + err_exit: return err; } @@ -234,6 +685,8 @@ void aq_ptp_ring_stop(struct aq_nic_s *aq_nic) aq_nic->aq_hw_ops->hw_ring_rx_stop(aq_nic->aq_hw, &aq_ptp->ptp_rx); aq_nic->aq_hw_ops->hw_ring_rx_stop(aq_nic->aq_hw, &aq_ptp->hwts_rx); + + napi_disable(&aq_ptp->napi); } void aq_ptp_ring_deinit(struct aq_nic_s *aq_nic) @@ -306,6 +759,12 @@ int aq_ptp_ring_alloc(struct aq_nic_s *aq_nic) goto err_exit_hwts_rx; } + aq_ptp->ptp_ring_param.vec_idx = aq_ptp->idx_vector; + aq_ptp->ptp_ring_param.cpu = aq_ptp->ptp_ring_param.vec_idx + + aq_nic_get_cfg(aq_nic)->aq_rss.base_cpu_number; + cpumask_set_cpu(aq_ptp->ptp_ring_param.cpu, + &aq_ptp->ptp_ring_param.affinity_mask); + return 0; err_exit_hwts_rx: @@ -347,6 +806,60 @@ static struct ptp_clock_info aq_ptp_clock = { .pin_config = NULL, }; +#define ptp_offset_init(__idx, __mbps, __egress, __ingress) do { \ + ptp_offset[__idx].mbps = (__mbps); \ + ptp_offset[__idx].egress = (__egress); \ + ptp_offset[__idx].ingress = (__ingress); } \ + while (0) + +static void aq_ptp_offset_init_from_fw(const struct hw_aq_ptp_offset *offsets) +{ + int i; + + /* Load offsets for PTP */ + for (i = 0; i < ARRAY_SIZE(ptp_offset); i++) { + switch (i) { + /* 100M */ + case ptp_offset_idx_100: + ptp_offset_init(i, 100, + offsets->egress_100, + offsets->ingress_100); + break; + /* 1G */ + case ptp_offset_idx_1000: + ptp_offset_init(i, 1000, + offsets->egress_1000, + offsets->ingress_1000); + break; + /* 2.5G */ + case ptp_offset_idx_2500: + ptp_offset_init(i, 2500, + offsets->egress_2500, + offsets->ingress_2500); + break; + /* 5G */ + case ptp_offset_idx_5000: + ptp_offset_init(i, 5000, + offsets->egress_5000, + offsets->ingress_5000); + break; + /* 10G */ + case ptp_offset_idx_10000: + ptp_offset_init(i, 10000, + offsets->egress_10000, + offsets->ingress_10000); + break; + } + } +} + +static void aq_ptp_offset_init(const struct hw_aq_ptp_offset *offsets) +{ + memset(ptp_offset, 0, sizeof(ptp_offset)); + + aq_ptp_offset_init_from_fw(offsets); +} + void aq_ptp_clock_init(struct aq_nic_s *aq_nic) { struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp; @@ -380,6 +893,8 @@ int aq_ptp_init(struct aq_nic_s *aq_nic, unsigned int idx_vec) return 0; } + aq_ptp_offset_init(&mbox.info.ptp_offset); + aq_ptp = kzalloc(sizeof(*aq_ptp), GFP_KERNEL); if (!aq_ptp) { err = -ENOMEM; @@ -399,6 +914,15 @@ int aq_ptp_init(struct aq_nic_s *aq_nic, unsigned int idx_vec) goto err_exit; } aq_ptp->ptp_clock = clock; + aq_ptp_tx_timeout_init(&aq_ptp->ptp_tx_timeout); + + atomic_set(&aq_ptp->offset_egress, 0); + atomic_set(&aq_ptp->offset_ingress, 0); + + netif_napi_add(aq_nic_get_ndev(aq_nic), &aq_ptp->napi, + aq_ptp_poll, AQ_CFG_NAPI_WEIGHT); + + aq_ptp->idx_vector = idx_vec; aq_nic->aq_ptp = aq_ptp; @@ -439,6 +963,12 @@ void aq_ptp_free(struct aq_nic_s *aq_nic) aq_nic->aq_fw_ops->enable_ptp(aq_nic->aq_hw, 0); mutex_unlock(&aq_nic->fwreq_mutex); + netif_napi_del(&aq_ptp->napi); kfree(aq_ptp); aq_nic->aq_ptp = NULL; } + +struct ptp_clock *aq_ptp_get_ptp_clock(struct aq_ptp_s *aq_ptp) +{ + return aq_ptp->ptp_clock; +} -- cgit v1.2.3-59-g8ed1b