LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: kerneljasonxing@gmail.com
To: jesse.brandeburg@intel.com, anthony.l.nguyen@intel.com,
	davem@davemloft.net, kuba@kernel.org, ast@kernel.org,
	daniel@iogearbox.net, hawk@kernel.org, john.fastabend@gmail.com,
	andrii@kernel.org, kafai@fb.com, songliubraving@fb.com,
	yhs@fb.com, kpsingh@kernel.org
Cc: intel-wired-lan@lists.osuosl.org, netdev@vger.kernel.org,
	linux-kernel@vger.kernel.org, bpf@vger.kernel.org,
	kerneljasonxing@gmail.com, Jason Xing <xingwanli@kuaishou.com>,
	Shujin Li <lishujin@kuaishou.com>
Subject: [PATCH v3] ixgbe: let the xdpdrv work with more than 64 cpus
Date: Thu, 26 Aug 2021 22:01:01 +0800	[thread overview]
Message-ID: <20210826140101.7944-1-kerneljasonxing@gmail.com> (raw)
In-Reply-To: <CAL+tcoCrOc1L+Y_SeScYJXjn542GYvu9n7EMhN_75h-P4FQFoQ@mail.gmail.com>

From: Jason Xing <xingwanli@kuaishou.com>

Originally, ixgbe driver doesn't allow the mounting of xdpdrv if the
server is equipped with more than 64 cpus online. So it turns out that
the loading of xdpdrv causes the "NOMEM" failure.

Actually, we can adjust the algorithm and then make it work, which has
no harm at all, only if we set the maxmium number of xdp queues.

Fixes: 33fdc82f08 ("ixgbe: add support for XDP_TX action")
Co-developed-by: Shujin Li <lishujin@kuaishou.com>
Signed-off-by: Shujin Li <lishujin@kuaishou.com>
Signed-off-by: Jason Xing <xingwanli@kuaishou.com>
---
 drivers/net/ethernet/intel/ixgbe/ixgbe.h           | 15 ++++-
 drivers/net/ethernet/intel/ixgbe/ixgbe_lib.c       |  9 ++-
 drivers/net/ethernet/intel/ixgbe/ixgbe_main.c      | 64 ++++++++++++++++------
 .../net/ethernet/intel/ixgbe/ixgbe_txrx_common.h   |  1 +
 drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c       |  9 +--
 5 files changed, 73 insertions(+), 25 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe.h b/drivers/net/ethernet/intel/ixgbe/ixgbe.h
index a604552..5f7f181 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe.h
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe.h
@@ -82,6 +82,8 @@
 #define IXGBE_2K_TOO_SMALL_WITH_PADDING \
 ((NET_SKB_PAD + IXGBE_RXBUFFER_1536) > SKB_WITH_OVERHEAD(IXGBE_RXBUFFER_2K))
 
+DECLARE_STATIC_KEY_FALSE(ixgbe_xdp_locking_key);
+
 static inline int ixgbe_compute_pad(int rx_buf_len)
 {
 	int page_size, pad_size;
@@ -351,6 +353,7 @@ struct ixgbe_ring {
 	};
 	u16 rx_offset;
 	struct xdp_rxq_info xdp_rxq;
+	spinlock_t tx_lock;	/* used in XDP mode */
 	struct xsk_buff_pool *xsk_pool;
 	u16 ring_idx;		/* {rx,tx,xdp}_ring back reference idx */
 	u16 rx_buf_len;
@@ -375,7 +378,7 @@ enum ixgbe_ring_f_enum {
 #define IXGBE_MAX_FCOE_INDICES		8
 #define MAX_RX_QUEUES			(IXGBE_MAX_FDIR_INDICES + 1)
 #define MAX_TX_QUEUES			(IXGBE_MAX_FDIR_INDICES + 1)
-#define MAX_XDP_QUEUES			(IXGBE_MAX_FDIR_INDICES + 1)
+#define IXGBE_MAX_XDP_QS		(IXGBE_MAX_FDIR_INDICES + 1)
 #define IXGBE_MAX_L2A_QUEUES		4
 #define IXGBE_BAD_L2A_QUEUE		3
 #define IXGBE_MAX_MACVLANS		63
@@ -629,7 +632,7 @@ struct ixgbe_adapter {
 
 	/* XDP */
 	int num_xdp_queues;
-	struct ixgbe_ring *xdp_ring[MAX_XDP_QUEUES];
+	struct ixgbe_ring *xdp_ring[IXGBE_MAX_XDP_QS];
 	unsigned long *af_xdp_zc_qps; /* tracks AF_XDP ZC enabled rings */
 
 	/* TX */
@@ -772,6 +775,14 @@ struct ixgbe_adapter {
 #endif /* CONFIG_IXGBE_IPSEC */
 };
 
+static inline int ixgbe_determine_xdp_q_idx(int cpu)
+{
+	if (static_key_enabled(&ixgbe_xdp_locking_key))
+		return cpu % IXGBE_MAX_XDP_QS;
+	else
+		return cpu;
+}
+
 static inline u8 ixgbe_max_rss_indices(struct ixgbe_adapter *adapter)
 {
 	switch (adapter->hw.mac.type) {
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_lib.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_lib.c
index 0218f6c..884bf99 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_lib.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_lib.c
@@ -299,7 +299,10 @@ static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
 
 static int ixgbe_xdp_queues(struct ixgbe_adapter *adapter)
 {
-	return adapter->xdp_prog ? nr_cpu_ids : 0;
+	int queues;
+
+	queues = min_t(int, IXGBE_MAX_XDP_QS, num_online_cpus());
+	return adapter->xdp_prog ? queues : 0;
 }
 
 #define IXGBE_RSS_64Q_MASK	0x3F
@@ -947,6 +950,7 @@ static int ixgbe_alloc_q_vector(struct ixgbe_adapter *adapter,
 		ring->count = adapter->tx_ring_count;
 		ring->queue_index = xdp_idx;
 		set_ring_xdp(ring);
+		spin_lock_init(&ring->tx_lock);
 
 		/* assign ring to adapter */
 		WRITE_ONCE(adapter->xdp_ring[xdp_idx], ring);
@@ -1032,6 +1036,9 @@ static void ixgbe_free_q_vector(struct ixgbe_adapter *adapter, int v_idx)
 	adapter->q_vector[v_idx] = NULL;
 	__netif_napi_del(&q_vector->napi);
 
+	if (static_key_enabled(&ixgbe_xdp_locking_key))
+		static_branch_dec(&ixgbe_xdp_locking_key);
+
 	/*
 	 * after a call to __netif_napi_del() napi may still be used and
 	 * ixgbe_get_stats64() might access the rings on this vector,
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
index 14aea40..a878f40 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
@@ -165,6 +165,9 @@ static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
 MODULE_LICENSE("GPL v2");
 
+DEFINE_STATIC_KEY_FALSE(ixgbe_xdp_locking_key);
+EXPORT_SYMBOL(ixgbe_xdp_locking_key);
+
 static struct workqueue_struct *ixgbe_wq;
 
 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
@@ -2422,13 +2425,10 @@ static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 		xdp_do_flush_map();
 
 	if (xdp_xmit & IXGBE_XDP_TX) {
-		struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
+		int index = ixgbe_determine_xdp_q_idx(smp_processor_id());
+		struct ixgbe_ring *ring = adapter->xdp_ring[index];
 
-		/* Force memory writes to complete before letting h/w
-		 * know there are new descriptors to fetch.
-		 */
-		wmb();
-		writel(ring->next_to_use, ring->tail);
+		ixgbe_xdp_ring_update_tail_locked(ring);
 	}
 
 	u64_stats_update_begin(&rx_ring->syncp);
@@ -6320,7 +6320,7 @@ static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
 	if (ixgbe_init_rss_key(adapter))
 		return -ENOMEM;
 
-	adapter->af_xdp_zc_qps = bitmap_zalloc(MAX_XDP_QUEUES, GFP_KERNEL);
+	adapter->af_xdp_zc_qps = bitmap_zalloc(IXGBE_MAX_XDP_QS, GFP_KERNEL);
 	if (!adapter->af_xdp_zc_qps)
 		return -ENOMEM;
 
@@ -8539,21 +8539,32 @@ static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
 			struct xdp_frame *xdpf)
 {
-	struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
 	struct ixgbe_tx_buffer *tx_buffer;
 	union ixgbe_adv_tx_desc *tx_desc;
+	struct ixgbe_ring *ring;
 	u32 len, cmd_type;
 	dma_addr_t dma;
+	int index, ret;
 	u16 i;
 
 	len = xdpf->len;
 
-	if (unlikely(!ixgbe_desc_unused(ring)))
-		return IXGBE_XDP_CONSUMED;
+	index = ixgbe_determine_xdp_q_idx(smp_processor_id());
+	ring = adapter->xdp_ring[index];
+
+	if (static_branch_unlikely(&ixgbe_xdp_locking_key))
+		spin_lock(&ring->tx_lock);
+
+	if (unlikely(!ixgbe_desc_unused(ring))) {
+		ret = IXGBE_XDP_CONSUMED;
+		goto out;
+	}
 
 	dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
-	if (dma_mapping_error(ring->dev, dma))
-		return IXGBE_XDP_CONSUMED;
+	if (dma_mapping_error(ring->dev, dma)) {
+		ret = IXGBE_XDP_CONSUMED;
+		goto out;
+	}
 
 	/* record the location of the first descriptor for this packet */
 	tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
@@ -8590,7 +8601,11 @@ int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
 	tx_buffer->next_to_watch = tx_desc;
 	ring->next_to_use = i;
 
-	return IXGBE_XDP_TX;
+	ret = IXGBE_XDP_TX;
+out:
+	if (static_branch_unlikely(&ixgbe_xdp_locking_key))
+		spin_unlock(&ring->tx_lock);
+	return ret;
 }
 
 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
@@ -10130,8 +10145,13 @@ static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
 			return -EINVAL;
 	}
 
-	if (nr_cpu_ids > MAX_XDP_QUEUES)
+	/* if the number of cpus is much larger than the maximum of queues,
+	 * we should stop it and then return with NOMEM like before.
+	 */
+	if (num_online_cpus() > IXGBE_MAX_XDP_QS * 2)
 		return -ENOMEM;
+	else if (num_online_cpus() > IXGBE_MAX_XDP_QS)
+		static_branch_inc(&ixgbe_xdp_locking_key);
 
 	old_prog = xchg(&adapter->xdp_prog, prog);
 	need_reset = (!!prog != !!old_prog);
@@ -10195,12 +10215,22 @@ void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
 	writel(ring->next_to_use, ring->tail);
 }
 
+void ixgbe_xdp_ring_update_tail_locked(struct ixgbe_ring *ring)
+{
+	if (static_branch_unlikely(&ixgbe_xdp_locking_key))
+		spin_lock(&ring->tx_lock);
+	ixgbe_xdp_ring_update_tail(ring);
+	if (static_branch_unlikely(&ixgbe_xdp_locking_key))
+		spin_unlock(&ring->tx_lock);
+}
+
 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
 			  struct xdp_frame **frames, u32 flags)
 {
 	struct ixgbe_adapter *adapter = netdev_priv(dev);
 	struct ixgbe_ring *ring;
 	int nxmit = 0;
+	int index;
 	int i;
 
 	if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
@@ -10209,10 +10239,12 @@ static int ixgbe_xdp_xmit(struct net_device *dev, int n,
 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
 		return -EINVAL;
 
+	index = ixgbe_determine_xdp_q_idx(smp_processor_id());
+
 	/* During program transitions its possible adapter->xdp_prog is assigned
 	 * but ring has not been configured yet. In this case simply abort xmit.
 	 */
-	ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
+	ring = adapter->xdp_prog ? adapter->xdp_ring[index] : NULL;
 	if (unlikely(!ring))
 		return -ENXIO;
 
@@ -10230,7 +10262,7 @@ static int ixgbe_xdp_xmit(struct net_device *dev, int n,
 	}
 
 	if (unlikely(flags & XDP_XMIT_FLUSH))
-		ixgbe_xdp_ring_update_tail(ring);
+		ixgbe_xdp_ring_update_tail_locked(ring);
 
 	return nxmit;
 }
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_txrx_common.h b/drivers/net/ethernet/intel/ixgbe/ixgbe_txrx_common.h
index 2aeec78..f6426d9 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_txrx_common.h
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_txrx_common.h
@@ -23,6 +23,7 @@ void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
 void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
 		  struct sk_buff *skb);
 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring);
+void ixgbe_xdp_ring_update_tail_locked(struct ixgbe_ring *ring);
 void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter, u64 qmask);
 
 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring);
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c
index b1d22e4..82d00e4 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c
@@ -334,13 +334,10 @@ int ixgbe_clean_rx_irq_zc(struct ixgbe_q_vector *q_vector,
 		xdp_do_flush_map();
 
 	if (xdp_xmit & IXGBE_XDP_TX) {
-		struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
+		int index = ixgbe_determine_xdp_q_idx(smp_processor_id());
+		struct ixgbe_ring *ring = adapter->xdp_ring[index];
 
-		/* Force memory writes to complete before letting h/w
-		 * know there are new descriptors to fetch.
-		 */
-		wmb();
-		writel(ring->next_to_use, ring->tail);
+		ixgbe_xdp_ring_update_tail_locked(ring);
 	}
 
 	u64_stats_update_begin(&rx_ring->syncp);
-- 
1.8.3.1


  reply	other threads:[~2021-08-26 14:01 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-08-24 10:49 [PATCH] " kerneljasonxing
2021-08-24 13:32 ` Jesper Dangaard Brouer
2021-08-24 15:23   ` Jason Xing
2021-08-24 15:32     ` Maciej Fijalkowski
2021-08-25 11:59       ` Jason Xing
2021-08-25 12:02         ` [PATCH v2] " kerneljasonxing
2021-08-26  8:51           ` Maciej Fijalkowski
2021-08-26 10:56             ` Jason Xing
2021-08-26 14:01               ` kerneljasonxing [this message]
2021-08-26 14:04                 ` [PATCH v3] " Jason Xing

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=20210826140101.7944-1-kerneljasonxing@gmail.com \
    --to=kerneljasonxing@gmail.com \
    --cc=andrii@kernel.org \
    --cc=anthony.l.nguyen@intel.com \
    --cc=ast@kernel.org \
    --cc=bpf@vger.kernel.org \
    --cc=daniel@iogearbox.net \
    --cc=davem@davemloft.net \
    --cc=hawk@kernel.org \
    --cc=intel-wired-lan@lists.osuosl.org \
    --cc=jesse.brandeburg@intel.com \
    --cc=john.fastabend@gmail.com \
    --cc=kafai@fb.com \
    --cc=kpsingh@kernel.org \
    --cc=kuba@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=lishujin@kuaishou.com \
    --cc=netdev@vger.kernel.org \
    --cc=songliubraving@fb.com \
    --cc=xingwanli@kuaishou.com \
    --cc=yhs@fb.com \
    --subject='Re: [PATCH v3] ixgbe: let the xdpdrv work with more than 64 cpus' \
    /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

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).