| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAL+tcoBPZj3ixF8yXdEn7sYGzPHkBwFZWT3y6WQbcMghML42sA@mail.gmail.com>
Date: Mon, 30 Aug 2021 21:17:46 +0800
From: Jason Xing <kerneljasonxing@...il.com>
To: Jesse Brandeburg <jesse.brandeburg@...el.com>,
"Nguyen, Anthony L" <anthony.l.nguyen@...el.com>,
David Miller <davem@...emloft.net>, kuba@...nel.org,
ast@...nel.org, daniel@...earbox.net, hawk@...nel.org,
john.fastabend@...il.com, andrii@...nel.org, kafai@...com,
songliubraving@...com, yhs@...com, kpsingh@...nel.org
Cc: intel-wired-lan@...ts.osuosl.org, netdev <netdev@...r.kernel.org>,
LKML <linux-kernel@...r.kernel.org>, bpf@...r.kernel.org,
Jason Xing <xingwanli@...ishou.com>,
Shujin Li <lishujin@...ishou.com>
Subject: Re: [PATCH v5] ixgbe: let the xdpdrv work with more than 64 cpus
On Mon, Aug 30, 2021 at 8:49 PM <kerneljasonxing@...il.com> wrote:
>
> From: Jason Xing <xingwanli@...ishou.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 through
> mapping the current cpu to some xdp ring with the protect of @tx_lock.
>
> Here're some numbers before/after applying this patch with xdp-example
> loaded on the eth0X:
>
> Before After
> TCP_STREAM send-64 734.14 714.20
> TCP_STREAM send-128 1401.91 1395.05
> TCP_STREAM send-512 5311.67 5292.84
> TCP_STREAM send-1k 9277.40 9356.22 (not stable)
> TCP_RR send-1 22559.75 21844.22
> TCP_RR send-128 23169.54 22725.13
> TCP_RR send-512 21670.91 21412.56
>
Sorry, I need to update the full result:
As client with xdpdrv loaded (tx path):
Before After
TCP_STREAM send-64 734.14 714.20
TCP_STREAM send-128 1401.91 1395.05
TCP_STREAM send-512 5311.67 5292.84
TCP_STREAM send-1k 9277.40 9356.22 (not stable)
TCP_RR send-1 22559.75 21844.22
TCP_RR send-128 23169.54 22725.13
TCP_RR send-512 21670.91 21412.56
As server with xdpdrv loaded (rx path):
Before After
TCP_STREAM send-64 1416.49 1383.12
TCP_STREAM send-128 3141.49 3055.50
TCP_STREAM send-512 9488.73 9487.44
TCP_STREAM send-1k 9491.17 9356.22 (not stable)
TCP_RR send-1 23617.74 23601.60
Now, the numbers are complete:)
> I tested many times with different parameters combined through netperf.
> Though the result is not that accurate, I cannot see much influence on
> this patch. The static key is places on the hot path, but it actually
> shouldn't cause a huge regression theoretically.
>
> Fixes: 33fdc82f08 ("ixgbe: add support for XDP_TX action")
> Co-developed-by: Shujin Li <lishujin@...ishou.com>
> Signed-off-by: Shujin Li <lishujin@...ishou.com>
> Signed-off-by: Jason Xing <xingwanli@...ishou.com>
> ---
> v5:
> - Change back to nr_cpu_ids (Eric)
> - Add the basic net performance data (Jason)
>
> v4:
> - Update the wrong commit messages. (Jason)
>
> v3:
> - Change nr_cpu_ids to num_online_cpus() (Maciej)
> - Rename MAX_XDP_QUEUES to IXGBE_MAX_XDP_QS (Maciej)
> - Rename ixgbe_determine_xdp_cpu() to ixgbe_determine_xdp_q_idx() (Maciej)
> - Wrap ixgbe_xdp_ring_update_tail() with lock into one function (Maciej)
>
> v2:
> - Adjust cpu id in ixgbe_xdp_xmit(). (Jesper)
> - Add a fallback path. (Maciej)
> - Adjust other parts related to xdp ring.
> ---
> 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..86b1116 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, nr_cpu_ids);
> + 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..bec29f5 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 (nr_cpu_ids > IXGBE_MAX_XDP_QS * 2)
> return -ENOMEM;
> + else if (nr_cpu_ids > 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
>
Powered by blists - more mailing lists