nfp: bpf: add hardware bpf offload

Add hardware bpf offload on our smart NICs.  Detect if
capable firmware is loaded and use it to load the code JITed
with just added translator onto programmable engines.

This commit only supports offloading cls_bpf in legacy mode
(non-direct action).
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/netronome/nfp/Makefile

@@ -3,6 +3,7 @@ obj-$(CONFIG_NFP_NETVF)	+= nfp_netvf.o
 nfp_netvf-objs := \
 	    nfp_net_common.o \
 	    nfp_net_ethtool.o \
+	    nfp_net_offload.o \
 	    nfp_netvf_main.o
 
 ifeq ($(CONFIG_BPF_SYSCALL),y)

drivers/net/ethernet/netronome/nfp/nfp_net.h

@@ -220,7 +220,7 @@ struct nfp_net_tx_ring {
 #define PCIE_DESC_RX_I_TCP_CSUM_OK	cpu_to_le16(BIT(11))
 #define PCIE_DESC_RX_I_UDP_CSUM		cpu_to_le16(BIT(10))
 #define PCIE_DESC_RX_I_UDP_CSUM_OK	cpu_to_le16(BIT(9))
-#define PCIE_DESC_RX_SPARE		cpu_to_le16(BIT(8))
+#define PCIE_DESC_RX_BPF		cpu_to_le16(BIT(8))
 #define PCIE_DESC_RX_EOP		cpu_to_le16(BIT(7))
 #define PCIE_DESC_RX_IP4_CSUM		cpu_to_le16(BIT(6))
 #define PCIE_DESC_RX_IP4_CSUM_OK	cpu_to_le16(BIT(5))
@@ -413,6 +413,7 @@ static inline bool nfp_net_fw_ver_eq(struct nfp_net_fw_version *fw_ver,
  * @is_vf:              Is the driver attached to a VF?
  * @is_nfp3200:         Is the driver for a NFP-3200 card?
  * @fw_loaded:          Is the firmware loaded?
+ * @bpf_offload_skip_sw:  Offloaded BPF program will not be rerun by cls_bpf
  * @ctrl:               Local copy of the control register/word.
  * @fl_bufsz:           Currently configured size of the freelist buffers
  * @rx_offset:		Offset in the RX buffers where packet data starts
@@ -473,6 +474,7 @@ struct nfp_net {
 	unsigned is_vf:1;
 	unsigned is_nfp3200:1;
 	unsigned fw_loaded:1;
+	unsigned bpf_offload_skip_sw:1;
 
 	u32 ctrl;
 	u32 fl_bufsz;
@@ -561,12 +563,28 @@ struct nfp_net {
 /* Functions to read/write from/to a BAR
  * Performs any endian conversion necessary.
  */
+static inline u16 nn_readb(struct nfp_net *nn, int off)
+{
+	return readb(nn->ctrl_bar + off);
+}
+
 static inline void nn_writeb(struct nfp_net *nn, int off, u8 val)
 {
 	writeb(val, nn->ctrl_bar + off);
 }
 
-/* NFP-3200 can't handle 16-bit accesses too well - hence no readw/writew */
+/* NFP-3200 can't handle 16-bit accesses too well */
+static inline u16 nn_readw(struct nfp_net *nn, int off)
+{
+	WARN_ON_ONCE(nn->is_nfp3200);
+	return readw(nn->ctrl_bar + off);
+}
+
+static inline void nn_writew(struct nfp_net *nn, int off, u16 val)
+{
+	WARN_ON_ONCE(nn->is_nfp3200);
+	writew(val, nn->ctrl_bar + off);
+}
 
 static inline u32 nn_readl(struct nfp_net *nn, int off)
 {
@@ -757,4 +775,8 @@ static inline void nfp_net_debugfs_adapter_del(struct nfp_net *nn)
 }
 #endif /* CONFIG_NFP_NET_DEBUG */
 
+int
+nfp_net_bpf_offload(struct nfp_net *nn, u32 handle, __be16 proto,
+		    struct tc_cls_bpf_offload *cls_bpf);
+
 #endif /* _NFP_NET_H_ */

drivers/net/ethernet/netronome/nfp/nfp_net_common.c

@@ -60,6 +60,7 @@
 
 #include <linux/ktime.h>
 
+#include <net/pkt_cls.h>
 #include <net/vxlan.h>
 
 #include "nfp_net_ctrl.h"
@@ -2382,6 +2383,31 @@ static struct rtnl_link_stats64 *nfp_net_stat64(struct net_device *netdev,
 	return stats;
 }
 
+static bool nfp_net_ebpf_capable(struct nfp_net *nn)
+{
+	if (nn->cap & NFP_NET_CFG_CTRL_BPF &&
+	    nn_readb(nn, NFP_NET_CFG_BPF_ABI) == NFP_NET_BPF_ABI)
+		return true;
+	return false;
+}
+
+static int
+nfp_net_setup_tc(struct net_device *netdev, u32 handle, __be16 proto,
+		 struct tc_to_netdev *tc)
+{
+	struct nfp_net *nn = netdev_priv(netdev);
+
+	if (TC_H_MAJ(handle) != TC_H_MAJ(TC_H_INGRESS))
+		return -ENOTSUPP;
+	if (proto != htons(ETH_P_ALL))
+		return -ENOTSUPP;
+
+	if (tc->type == TC_SETUP_CLSBPF && nfp_net_ebpf_capable(nn))
+		return nfp_net_bpf_offload(nn, handle, proto, tc->cls_bpf);
+
+	return -EINVAL;
+}
+
 static int nfp_net_set_features(struct net_device *netdev,
 				netdev_features_t features)
 {
@@ -2436,6 +2462,11 @@ static int nfp_net_set_features(struct net_device *netdev,
 			new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER;
 	}
 
+	if (changed & NETIF_F_HW_TC && nn->ctrl & NFP_NET_CFG_CTRL_BPF) {
+		nn_err(nn, "Cannot disable HW TC offload while in use\n");
+		return -EBUSY;
+	}
+
 	nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n",
 	       netdev->features, features, changed);
 
@@ -2585,6 +2616,7 @@ static const struct net_device_ops nfp_net_netdev_ops = {
 	.ndo_stop		= nfp_net_netdev_close,
 	.ndo_start_xmit		= nfp_net_tx,
 	.ndo_get_stats64	= nfp_net_stat64,
+	.ndo_setup_tc		= nfp_net_setup_tc,
 	.ndo_tx_timeout		= nfp_net_tx_timeout,
 	.ndo_set_rx_mode	= nfp_net_set_rx_mode,
 	.ndo_change_mtu		= nfp_net_change_mtu,
@@ -2610,7 +2642,7 @@ void nfp_net_info(struct nfp_net *nn)
 		nn->fw_ver.resv, nn->fw_ver.class,
 		nn->fw_ver.major, nn->fw_ver.minor,
 		nn->max_mtu);
-	nn_info(nn, "CAP: %#x %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
+	nn_info(nn, "CAP: %#x %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
 		nn->cap,
 		nn->cap & NFP_NET_CFG_CTRL_PROMISC  ? "PROMISC "  : "",
 		nn->cap & NFP_NET_CFG_CTRL_L2BC     ? "L2BCFILT " : "",
@@ -2627,7 +2659,8 @@ void nfp_net_info(struct nfp_net *nn)
 		nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "",
 		nn->cap & NFP_NET_CFG_CTRL_IRQMOD   ? "IRQMOD "   : "",
 		nn->cap & NFP_NET_CFG_CTRL_VXLAN    ? "VXLAN "    : "",
-		nn->cap & NFP_NET_CFG_CTRL_NVGRE    ? "NVGRE "	  : "");
+		nn->cap & NFP_NET_CFG_CTRL_NVGRE    ? "NVGRE "	  : "",
+		nfp_net_ebpf_capable(nn)            ? "BPF "	  : "");
 }
 
 /**
@@ -2795,6 +2828,9 @@ int nfp_net_netdev_init(struct net_device *netdev)
 
 	netdev->features = netdev->hw_features;
 
+	if (nfp_net_ebpf_capable(nn))
+		netdev->hw_features |= NETIF_F_HW_TC;
+
 	/* Advertise but disable TSO by default. */
 	netdev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
 

drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h

@@ -123,6 +123,7 @@
 #define   NFP_NET_CFG_CTRL_L2SWITCH_LOCAL (0x1 << 23) /* Switch to local */
 #define   NFP_NET_CFG_CTRL_VXLAN	  (0x1 << 24) /* VXLAN tunnel support */
 #define   NFP_NET_CFG_CTRL_NVGRE	  (0x1 << 25) /* NVGRE tunnel support */
+#define   NFP_NET_CFG_CTRL_BPF		  (0x1 << 27) /* BPF offload capable */
 #define NFP_NET_CFG_UPDATE              0x0004
 #define   NFP_NET_CFG_UPDATE_GEN          (0x1 <<  0) /* General update */
 #define   NFP_NET_CFG_UPDATE_RING         (0x1 <<  1) /* Ring config change */
@@ -134,6 +135,7 @@
 #define   NFP_NET_CFG_UPDATE_RESET        (0x1 <<  7) /* Update due to FLR */
 #define   NFP_NET_CFG_UPDATE_IRQMOD       (0x1 <<  8) /* IRQ mod change */
 #define   NFP_NET_CFG_UPDATE_VXLAN	  (0x1 <<  9) /* VXLAN port change */
+#define   NFP_NET_CFG_UPDATE_BPF	  (0x1 << 10) /* BPF program load */
 #define   NFP_NET_CFG_UPDATE_ERR          (0x1 << 31) /* A error occurred */
 #define NFP_NET_CFG_TXRS_ENABLE         0x0008
 #define NFP_NET_CFG_RXRS_ENABLE         0x0010
@@ -196,10 +198,37 @@
 #define NFP_NET_CFG_VXLAN_SZ		  0x0008
 
 /**
- * 64B reserved for future use (0x0080 - 0x00c0)
+ * NFP6000 - BPF section
+ * @NFP_NET_CFG_BPF_ABI:	BPF ABI version
+ * @NFP_NET_CFG_BPF_CAP:	BPF capabilities
+ * @NFP_NET_CFG_BPF_MAX_LEN:	Maximum size of JITed BPF code in bytes
+ * @NFP_NET_CFG_BPF_START:	Offset at which BPF will be loaded
+ * @NFP_NET_CFG_BPF_DONE:	Offset to jump to on exit
+ * @NFP_NET_CFG_BPF_STACK_SZ:	Total size of stack area in 64B chunks
+ * @NFP_NET_CFG_BPF_INL_MTU:	Packet data split offset in 64B chunks
+ * @NFP_NET_CFG_BPF_SIZE:	Size of the JITed BPF code in instructions
+ * @NFP_NET_CFG_BPF_ADDR:	DMA address of the buffer with JITed BPF code
  */
-#define NFP_NET_CFG_RESERVED            0x0080
-#define NFP_NET_CFG_RESERVED_SZ         0x0040
+#define NFP_NET_CFG_BPF_ABI		0x0080
+#define   NFP_NET_BPF_ABI		1
+#define NFP_NET_CFG_BPF_CAP		0x0081
+#define   NFP_NET_BPF_CAP_RELO		(1 << 0) /* seamless reload */
+#define NFP_NET_CFG_BPF_MAX_LEN		0x0082
+#define NFP_NET_CFG_BPF_START		0x0084
+#define NFP_NET_CFG_BPF_DONE		0x0086
+#define NFP_NET_CFG_BPF_STACK_SZ	0x0088
+#define NFP_NET_CFG_BPF_INL_MTU		0x0089
+#define NFP_NET_CFG_BPF_SIZE		0x008e
+#define NFP_NET_CFG_BPF_ADDR		0x0090
+#define   NFP_NET_CFG_BPF_CFG_8CTX	(1 << 0) /* 8ctx mode */
+#define   NFP_NET_CFG_BPF_CFG_MASK	7ULL
+#define   NFP_NET_CFG_BPF_ADDR_MASK	(~NFP_NET_CFG_BPF_CFG_MASK)
+
+/**
+ * 40B reserved for future use (0x0098 - 0x00c0)
+ */
+#define NFP_NET_CFG_RESERVED            0x0098
+#define NFP_NET_CFG_RESERVED_SZ         0x0028
 
 /**
  * RSS configuration (0x0100 - 0x01ac):
@@ -303,6 +332,15 @@
 #define NFP_NET_CFG_STATS_TX_MC_FRAMES  (NFP_NET_CFG_STATS_BASE + 0x80)
 #define NFP_NET_CFG_STATS_TX_BC_FRAMES  (NFP_NET_CFG_STATS_BASE + 0x88)
 
+#define NFP_NET_CFG_STATS_APP0_FRAMES	(NFP_NET_CFG_STATS_BASE + 0x90)
+#define NFP_NET_CFG_STATS_APP0_BYTES	(NFP_NET_CFG_STATS_BASE + 0x98)
+#define NFP_NET_CFG_STATS_APP1_FRAMES	(NFP_NET_CFG_STATS_BASE + 0xa0)
+#define NFP_NET_CFG_STATS_APP1_BYTES	(NFP_NET_CFG_STATS_BASE + 0xa8)
+#define NFP_NET_CFG_STATS_APP2_FRAMES	(NFP_NET_CFG_STATS_BASE + 0xb0)
+#define NFP_NET_CFG_STATS_APP2_BYTES	(NFP_NET_CFG_STATS_BASE + 0xb8)
+#define NFP_NET_CFG_STATS_APP3_FRAMES	(NFP_NET_CFG_STATS_BASE + 0xc0)
+#define NFP_NET_CFG_STATS_APP3_BYTES	(NFP_NET_CFG_STATS_BASE + 0xc8)
+
 /**
  * Per ring stats (0x1000 - 0x1800)
  * options, 64bit per entry

drivers/net/ethernet/netronome/nfp/nfp_net_offload.c

+/*
+ * Copyright (C) 2016 Netronome Systems, Inc.
+ *
+ * This software is dual licensed under the GNU General License Version 2,
+ * June 1991 as shown in the file COPYING in the top-level directory of this
+ * source tree or the BSD 2-Clause License provided below.  You have the
+ * option to license this software under the complete terms of either license.
+ *
+ * The BSD 2-Clause License:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      1. Redistributions of source code must retain the above
+ *         copyright notice, this list of conditions and the following
+ *         disclaimer.
+ *
+ *      2. Redistributions in binary form must reproduce the above
+ *         copyright notice, this list of conditions and the following
+ *         disclaimer in the documentation and/or other materials
+ *         provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * nfp_net_offload.c
+ * Netronome network device driver: TC offload functions for PF and VF
+ */
+
+#include <linux/kernel.h>
+#include <linux/netdevice.h>
+#include <linux/pci.h>
+#include <linux/jiffies.h>
+#include <linux/timer.h>
+#include <linux/list.h>
+
+#include <net/pkt_cls.h>
+#include <net/tc_act/tc_gact.h>
+#include <net/tc_act/tc_mirred.h>
+
+#include "nfp_bpf.h"
+#include "nfp_net_ctrl.h"
+#include "nfp_net.h"
+
+static int
+nfp_net_bpf_get_act(struct nfp_net *nn, struct tc_cls_bpf_offload *cls_bpf)
+{
+	const struct tc_action *a;
+	LIST_HEAD(actions);
+
+	/* TC direct action */
+	if (cls_bpf->exts_integrated)
+		return -ENOTSUPP;
+
+	/* TC legacy mode */
+	if (!tc_single_action(cls_bpf->exts))
+		return -ENOTSUPP;
+
+	tcf_exts_to_list(cls_bpf->exts, &actions);
+	list_for_each_entry(a, &actions, list) {
+		if (is_tcf_gact_shot(a))
+			return NN_ACT_TC_DROP;
+	}
+
+	return -ENOTSUPP;
+}
+
+static int
+nfp_net_bpf_offload_prepare(struct nfp_net *nn,
+			    struct tc_cls_bpf_offload *cls_bpf,
+			    struct nfp_bpf_result *res,
+			    void **code, dma_addr_t *dma_addr, u16 max_instr)
+{
+	unsigned int code_sz = max_instr * sizeof(u64);
+	enum nfp_bpf_action_type act;
+	u16 start_off, done_off;
+	unsigned int max_mtu;
+	int ret;
+
+	ret = nfp_net_bpf_get_act(nn, cls_bpf);
+	if (ret < 0)
+		return ret;
+	act = ret;
+
+	max_mtu = nn_readb(nn, NFP_NET_CFG_BPF_INL_MTU) * 64 - 32;
+	if (max_mtu < nn->netdev->mtu) {
+		nn_info(nn, "BPF offload not supported with MTU larger than HW packet split boundary\n");
+		return -ENOTSUPP;
+	}
+
+	start_off = nn_readw(nn, NFP_NET_CFG_BPF_START);
+	done_off = nn_readw(nn, NFP_NET_CFG_BPF_DONE);
+
+	*code = dma_zalloc_coherent(&nn->pdev->dev, code_sz, dma_addr,
+				    GFP_KERNEL);
+	if (!*code)
+		return -ENOMEM;
+
+	ret = nfp_bpf_jit(cls_bpf->prog, *code, act, start_off, done_off,
+			  max_instr, res);
+	if (ret)
+		goto out;
+
+	return 0;
+
+out:
+	dma_free_coherent(&nn->pdev->dev, code_sz, *code, *dma_addr);
+	return ret;
+}
+
+static void
+nfp_net_bpf_load_and_start(struct nfp_net *nn, u32 tc_flags,
+			   void *code, dma_addr_t dma_addr,
+			   unsigned int code_sz, unsigned int n_instr,
+			   bool dense_mode)
+{
+	u64 bpf_addr = dma_addr;
+	int err;
+
+	nn->bpf_offload_skip_sw = !!(tc_flags & TCA_CLS_FLAGS_SKIP_SW);
+
+	if (dense_mode)
+		bpf_addr |= NFP_NET_CFG_BPF_CFG_8CTX;
+
+	nn_writew(nn, NFP_NET_CFG_BPF_SIZE, n_instr);
+	nn_writeq(nn, NFP_NET_CFG_BPF_ADDR, bpf_addr);
+
+	/* Load up the JITed code */
+	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_BPF);
+	if (err)
+		nn_err(nn, "FW command error while loading BPF: %d\n", err);
+
+	/* Enable passing packets through BPF function */
+	nn->ctrl |= NFP_NET_CFG_CTRL_BPF;
+	nn_writel(nn, NFP_NET_CFG_CTRL, nn->ctrl);
+	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
+	if (err)
+		nn_err(nn, "FW command error while enabling BPF: %d\n", err);
+
+	dma_free_coherent(&nn->pdev->dev, code_sz, code, dma_addr);
+}
+
+static int nfp_net_bpf_stop(struct nfp_net *nn)
+{
+	if (!(nn->ctrl & NFP_NET_CFG_CTRL_BPF))
+		return 0;
+
+	nn->ctrl &= ~NFP_NET_CFG_CTRL_BPF;
+	nn_writel(nn, NFP_NET_CFG_CTRL, nn->ctrl);
+
+	nn->bpf_offload_skip_sw = 0;
+
+	return nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
+}
+
+int
+nfp_net_bpf_offload(struct nfp_net *nn, u32 handle, __be16 proto,
+		    struct tc_cls_bpf_offload *cls_bpf)
+{
+	struct nfp_bpf_result res;
+	dma_addr_t dma_addr;
+	u16 max_instr;
+	void *code;
+	int err;
+
+	max_instr = nn_readw(nn, NFP_NET_CFG_BPF_MAX_LEN);
+
+	switch (cls_bpf->command) {
+	case TC_CLSBPF_REPLACE:
+		/* There is nothing stopping us from implementing seamless
+		 * replace but the simple method of loading I adopted in
+		 * the firmware does not handle atomic replace (i.e. we have to
+		 * stop the BPF offload and re-enable it).  Leaking-in a few
+		 * frames which didn't have BPF applied in the hardware should
+		 * be fine if software fallback is available, though.
+		 */
+		if (nn->bpf_offload_skip_sw)
+			return -EBUSY;
+
+		err = nfp_net_bpf_offload_prepare(nn, cls_bpf, &res, &code,
+						  &dma_addr, max_instr);
+		if (err)
+			return err;
+
+		nfp_net_bpf_stop(nn);
+		nfp_net_bpf_load_and_start(nn, cls_bpf->gen_flags, code,
+					   dma_addr, max_instr * sizeof(u64),
+					   res.n_instr, res.dense_mode);
+		return 0;
+
+	case TC_CLSBPF_ADD:
+		if (nn->ctrl & NFP_NET_CFG_CTRL_BPF)
+			return -EBUSY;
+
+		err = nfp_net_bpf_offload_prepare(nn, cls_bpf, &res, &code,
+						  &dma_addr, max_instr);
+		if (err)
+			return err;
+
+		nfp_net_bpf_load_and_start(nn, cls_bpf->gen_flags, code,
+					   dma_addr, max_instr * sizeof(u64),
+					   res.n_instr, res.dense_mode);
+		return 0;
+
+	case TC_CLSBPF_DESTROY:
+		return nfp_net_bpf_stop(nn);
+
+	default:
+		return -ENOTSUPP;
+	}
+}