[PATCHv5 net-next 00/15] BPF hardware offload (cls_bpf for now)

[PATCHv5 net-next 00/15] BPF hardware offload (cls_bpf for now)

[Jakub Kicinski]

Hi!

Dave, this set depends on bitfield.h which is sitting in the
pull request from Kalle.  Please pull wireless-drivers-next 
before applying.

v5:
 - fix names of guard defines in bpf_verfier.h.
v4:
 - rename parser -> analyzer;
 - reorganize the analyzer patches a bit;
 - use bitfield.h directly.

--- merge blurb:
In the last year a lot of progress have been made on offloading
simpler TC classifiers.  There is also growing interest in using
BPF for generic high-speed packet processing in the kernel.
It seems beneficial to tie those two trends together and think
about hardware offloads of BPF programs.  This patch set presents
such offload to Netronome smart NICs.  cls_bpf is extended with
hardware offload capabilities and NFP driver gets a JIT translator
which in presence of capable firmware can be used to offload
the BPF program onto the card.

BPF JIT implementation is not 100% complete (e.g. missing instructions)
but it is functional.  Encouragingly it should be possible to
offload most (if not all) advanced BPF features onto the NIC - 
including packet modification, maps, tunnel encap/decap etc.

Example of basic tests I used:
  __section_cls_entry
  int cls_entry(struct __sk_buff *skb)
  {
	if (load_byte(skb, 0) != 0x0)
		return 0;

	if (load_byte(skb, 4) != 0x1)
		return 0;

	skb->mark = 0xcafe;

	if (load_byte(skb, 50) != 0xff)
		return 0;

	return ~0U;
  }

Above code can be compiled with Clang and loaded like this:
# ethtool -K p1p1 hw-tc-offload on
# tc qdisc add dev p1p1 ingress
# tc filter add dev p1p1 parent ffff:  bpf obj prog.o action drop

This set implements the basic transparent offload, the skip_{sw,hw}
flags and reporting statistics for cls_bpf.

Jakub Kicinski (15):
  net: cls_bpf: add hardware offload
  net: cls_bpf: limit hardware offload by software-only flag
  net: cls_bpf: add support for marking filters as hardware-only
  bpf: don't (ab)use instructions to store state
  bpf: expose internal verfier structures
  bpf: enable non-core use of the verfier
  bpf: recognize 64bit immediate loads as consts
  nfp: add BPF to NFP code translator
  nfp: bpf: add hardware bpf offload
  net: cls_bpf: allow offloaded filters to update stats
  nfp: bpf: allow offloaded filters to update stats
  nfp: bpf: add packet marking support
  net: act_mirred: allow statistic updates from offloaded actions
  nfp: bpf: add support for legacy redirect action
  nfp: bpf: add offload of TC direct action mode

 drivers/net/ethernet/netronome/nfp/Makefile        |    7 +
 drivers/net/ethernet/netronome/nfp/nfp_asm.h       |  233 +++
 drivers/net/ethernet/netronome/nfp/nfp_bpf.h       |  212 +++
 drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c   | 1816 ++++++++++++++++++++
 .../net/ethernet/netronome/nfp/nfp_bpf_verifier.c  |  160 ++
 drivers/net/ethernet/netronome/nfp/nfp_net.h       |   47 +-
 .../net/ethernet/netronome/nfp/nfp_net_common.c    |  134 +-
 drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h  |   51 +-
 .../net/ethernet/netronome/nfp/nfp_net_ethtool.c   |   12 +
 .../net/ethernet/netronome/nfp/nfp_net_offload.c   |  291 ++++
 .../net/ethernet/netronome/nfp/nfp_netvf_main.c    |    2 +-
 include/linux/bpf_verifier.h                       |   89 +
 include/linux/netdevice.h                          |    2 +
 include/net/pkt_cls.h                              |   16 +
 include/uapi/linux/pkt_cls.h                       |    1 +
 kernel/bpf/verifier.c                              |  384 +++--
 net/sched/act_mirred.c                             |    8 +
 net/sched/cls_bpf.c                                |  117 +-
 18 files changed, 3376 insertions(+), 206 deletions(-)
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_asm.h
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_bpf.h
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_bpf_verifier.c
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
 create mode 100644 include/linux/bpf_verifier.h

-- 
1.9.1

[PATCHv5 net-next 01/15] net: cls_bpf: add hardware offload

[Jakub Kicinski]

This patch adds hardware offload capability to cls_bpf classifier,
similar to what have been done with U32 and flower.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
---
v3:
 - s/filter/prog/ in struct tc_cls_bpf_offload.
v2:
 - drop unnecessary WARN_ON;
 - reformat error handling a bit.
---
 include/linux/netdevice.h |  2 ++
 include/net/pkt_cls.h     | 14 ++++++++++
 net/sched/cls_bpf.c       | 70 +++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 86 insertions(+)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 2095b6ab3661..3c50db29a114 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -789,6 +789,7 @@ enum {
 	TC_SETUP_CLSU32,
 	TC_SETUP_CLSFLOWER,
 	TC_SETUP_MATCHALL,
+	TC_SETUP_CLSBPF,
 };
 
 struct tc_cls_u32_offload;
@@ -800,6 +801,7 @@ struct tc_to_netdev {
 		struct tc_cls_u32_offload *cls_u32;
 		struct tc_cls_flower_offload *cls_flower;
 		struct tc_cls_matchall_offload *cls_mall;
+		struct tc_cls_bpf_offload *cls_bpf;
 	};
 };
 
diff --git a/include/net/pkt_cls.h b/include/net/pkt_cls.h
index a459be5fe1c2..41e8071dff87 100644
--- a/include/net/pkt_cls.h
+++ b/include/net/pkt_cls.h
@@ -486,4 +486,18 @@ struct tc_cls_matchall_offload {
 	unsigned long cookie;
 };
 
+enum tc_clsbpf_command {
+	TC_CLSBPF_ADD,
+	TC_CLSBPF_REPLACE,
+	TC_CLSBPF_DESTROY,
+};
+
+struct tc_cls_bpf_offload {
+	enum tc_clsbpf_command command;
+	struct tcf_exts *exts;
+	struct bpf_prog *prog;
+	const char *name;
+	bool exts_integrated;
+};
+
 #endif
diff --git a/net/sched/cls_bpf.c b/net/sched/cls_bpf.c
index 1d92d4d3f222..3ca9502a881c 100644
--- a/net/sched/cls_bpf.c
+++ b/net/sched/cls_bpf.c
@@ -39,6 +39,7 @@ struct cls_bpf_prog {
 	struct list_head link;
 	struct tcf_result res;
 	bool exts_integrated;
+	bool offloaded;
 	struct tcf_exts exts;
 	u32 handle;
 	union {
@@ -137,6 +138,71 @@ static bool cls_bpf_is_ebpf(const struct cls_bpf_prog *prog)
 	return !prog->bpf_ops;
 }
 
+static int cls_bpf_offload_cmd(struct tcf_proto *tp, struct cls_bpf_prog *prog,
+			       enum tc_clsbpf_command cmd)
+{
+	struct net_device *dev = tp->q->dev_queue->dev;
+	struct tc_cls_bpf_offload bpf_offload = {};
+	struct tc_to_netdev offload;
+
+	offload.type = TC_SETUP_CLSBPF;
+	offload.cls_bpf = &bpf_offload;
+
+	bpf_offload.command = cmd;
+	bpf_offload.exts = &prog->exts;
+	bpf_offload.prog = prog->filter;
+	bpf_offload.name = prog->bpf_name;
+	bpf_offload.exts_integrated = prog->exts_integrated;
+
+	return dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle,
+					     tp->protocol, &offload);
+}
+
+static void cls_bpf_offload(struct tcf_proto *tp, struct cls_bpf_prog *prog,
+			    struct cls_bpf_prog *oldprog)
+{
+	struct net_device *dev = tp->q->dev_queue->dev;
+	struct cls_bpf_prog *obj = prog;
+	enum tc_clsbpf_command cmd;
+
+	if (oldprog && oldprog->offloaded) {
+		if (tc_should_offload(dev, tp, 0)) {
+			cmd = TC_CLSBPF_REPLACE;
+		} else {
+			obj = oldprog;
+			cmd = TC_CLSBPF_DESTROY;
+		}
+	} else {
+		if (!tc_should_offload(dev, tp, 0))
+			return;
+		cmd = TC_CLSBPF_ADD;
+	}
+
+	if (cls_bpf_offload_cmd(tp, obj, cmd))
+		return;
+
+	obj->offloaded = true;
+	if (oldprog)
+		oldprog->offloaded = false;
+}
+
+static void cls_bpf_stop_offload(struct tcf_proto *tp,
+				 struct cls_bpf_prog *prog)
+{
+	int err;
+
+	if (!prog->offloaded)
+		return;
+
+	err = cls_bpf_offload_cmd(tp, prog, TC_CLSBPF_DESTROY);
+	if (err) {
+		pr_err("Stopping hardware offload failed: %d\n", err);
+		return;
+	}
+
+	prog->offloaded = false;
+}
+
 static int cls_bpf_init(struct tcf_proto *tp)
 {
 	struct cls_bpf_head *head;
@@ -176,6 +242,7 @@ static int cls_bpf_delete(struct tcf_proto *tp, unsigned long arg)
 {
 	struct cls_bpf_prog *prog = (struct cls_bpf_prog *) arg;
 
+	cls_bpf_stop_offload(tp, prog);
 	list_del_rcu(&prog->link);
 	tcf_unbind_filter(tp, &prog->res);
 	call_rcu(&prog->rcu, __cls_bpf_delete_prog);
@@ -192,6 +259,7 @@ static bool cls_bpf_destroy(struct tcf_proto *tp, bool force)
 		return false;
 
 	list_for_each_entry_safe(prog, tmp, &head->plist, link) {
+		cls_bpf_stop_offload(tp, prog);
 		list_del_rcu(&prog->link);
 		tcf_unbind_filter(tp, &prog->res);
 		call_rcu(&prog->rcu, __cls_bpf_delete_prog);
@@ -413,6 +481,8 @@ static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
 	if (ret < 0)
 		goto errout;
 
+	cls_bpf_offload(tp, prog, oldprog);
+
 	if (oldprog) {
 		list_replace_rcu(&oldprog->link, &prog->link);
 		tcf_unbind_filter(tp, &oldprog->res);
-- 
1.9.1

[PATCHv5 net-next 02/15] net: cls_bpf: limit hardware offload by software-only flag

[Jakub Kicinski]

Add cls_bpf support for the TCA_CLS_FLAGS_SKIP_HW flag.
Unlike U32 and flower cls_bpf already has some netlink
flags defined.  Create a new attribute to be able to use
the same flag values as the above.

Unlike U32 and flower reject unknown flags.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
---
v3:
 - reject (instead of clear) unsupported flags;
 - fix error handling.
v2:
 - rename TCA_BPF_GEN_TCA_FLAGS -> TCA_BPF_FLAGS_GEN;
 - add comment about clearing unsupported flags;
 - validate flags after clearing unsupported.
---
 include/net/pkt_cls.h        |  1 +
 include/uapi/linux/pkt_cls.h |  1 +
 net/sched/cls_bpf.c          | 22 ++++++++++++++++++++--
 3 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/include/net/pkt_cls.h b/include/net/pkt_cls.h
index 41e8071dff87..57af9f3032ff 100644
--- a/include/net/pkt_cls.h
+++ b/include/net/pkt_cls.h
@@ -498,6 +498,7 @@ struct tc_cls_bpf_offload {
 	struct bpf_prog *prog;
 	const char *name;
 	bool exts_integrated;
+	u32 gen_flags;
 };
 
 #endif
diff --git a/include/uapi/linux/pkt_cls.h b/include/uapi/linux/pkt_cls.h
index 8915b61bbf83..8fd715f806a2 100644
--- a/include/uapi/linux/pkt_cls.h
+++ b/include/uapi/linux/pkt_cls.h
@@ -396,6 +396,7 @@ enum {
 	TCA_BPF_FD,
 	TCA_BPF_NAME,
 	TCA_BPF_FLAGS,
+	TCA_BPF_FLAGS_GEN,
 	__TCA_BPF_MAX,
 };
 
diff --git a/net/sched/cls_bpf.c b/net/sched/cls_bpf.c
index 3ca9502a881c..46af423a8a8f 100644
--- a/net/sched/cls_bpf.c
+++ b/net/sched/cls_bpf.c
@@ -27,6 +27,8 @@ MODULE_AUTHOR("Daniel Borkmann <dborkman@redhat.com>");
 MODULE_DESCRIPTION("TC BPF based classifier");
 
 #define CLS_BPF_NAME_LEN	256
+#define CLS_BPF_SUPPORTED_GEN_FLAGS		\
+	TCA_CLS_FLAGS_SKIP_HW
 
 struct cls_bpf_head {
 	struct list_head plist;
@@ -40,6 +42,7 @@ struct cls_bpf_prog {
 	struct tcf_result res;
 	bool exts_integrated;
 	bool offloaded;
+	u32 gen_flags;
 	struct tcf_exts exts;
 	u32 handle;
 	union {
@@ -55,6 +58,7 @@ struct cls_bpf_prog {
 static const struct nla_policy bpf_policy[TCA_BPF_MAX + 1] = {
 	[TCA_BPF_CLASSID]	= { .type = NLA_U32 },
 	[TCA_BPF_FLAGS]		= { .type = NLA_U32 },
+	[TCA_BPF_FLAGS_GEN]	= { .type = NLA_U32 },
 	[TCA_BPF_FD]		= { .type = NLA_U32 },
 	[TCA_BPF_NAME]		= { .type = NLA_NUL_STRING, .len = CLS_BPF_NAME_LEN },
 	[TCA_BPF_OPS_LEN]	= { .type = NLA_U16 },
@@ -153,6 +157,7 @@ static int cls_bpf_offload_cmd(struct tcf_proto *tp, struct cls_bpf_prog *prog,
 	bpf_offload.prog = prog->filter;
 	bpf_offload.name = prog->bpf_name;
 	bpf_offload.exts_integrated = prog->exts_integrated;
+	bpf_offload.gen_flags = prog->gen_flags;
 
 	return dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle,
 					     tp->protocol, &offload);
@@ -166,14 +171,14 @@ static void cls_bpf_offload(struct tcf_proto *tp, struct cls_bpf_prog *prog,
 	enum tc_clsbpf_command cmd;
 
 	if (oldprog && oldprog->offloaded) {
-		if (tc_should_offload(dev, tp, 0)) {
+		if (tc_should_offload(dev, tp, prog->gen_flags)) {
 			cmd = TC_CLSBPF_REPLACE;
 		} else {
 			obj = oldprog;
 			cmd = TC_CLSBPF_DESTROY;
 		}
 	} else {
-		if (!tc_should_offload(dev, tp, 0))
+		if (!tc_should_offload(dev, tp, prog->gen_flags))
 			return;
 		cmd = TC_CLSBPF_ADD;
 	}
@@ -369,6 +374,7 @@ static int cls_bpf_modify_existing(struct net *net, struct tcf_proto *tp,
 {
 	bool is_bpf, is_ebpf, have_exts = false;
 	struct tcf_exts exts;
+	u32 gen_flags = 0;
 	int ret;
 
 	is_bpf = tb[TCA_BPF_OPS_LEN] && tb[TCA_BPF_OPS];
@@ -393,8 +399,17 @@ static int cls_bpf_modify_existing(struct net *net, struct tcf_proto *tp,
 
 		have_exts = bpf_flags & TCA_BPF_FLAG_ACT_DIRECT;
 	}
+	if (tb[TCA_BPF_FLAGS_GEN]) {
+		gen_flags = nla_get_u32(tb[TCA_BPF_FLAGS_GEN]);
+		if (gen_flags & ~CLS_BPF_SUPPORTED_GEN_FLAGS ||
+		    !tc_flags_valid(gen_flags)) {
+			ret = -EINVAL;
+			goto errout;
+		}
+	}
 
 	prog->exts_integrated = have_exts;
+	prog->gen_flags = gen_flags;
 
 	ret = is_bpf ? cls_bpf_prog_from_ops(tb, prog) :
 		       cls_bpf_prog_from_efd(tb, prog, tp);
@@ -566,6 +581,9 @@ static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
 		bpf_flags |= TCA_BPF_FLAG_ACT_DIRECT;
 	if (bpf_flags && nla_put_u32(skb, TCA_BPF_FLAGS, bpf_flags))
 		goto nla_put_failure;
+	if (prog->gen_flags &&
+	    nla_put_u32(skb, TCA_BPF_FLAGS_GEN, prog->gen_flags))
+		goto nla_put_failure;
 
 	nla_nest_end(skb, nest);
 
-- 
1.9.1

[PATCHv5 net-next 03/15] net: cls_bpf: add support for marking filters as hardware-only

[Jakub Kicinski]

Add cls_bpf support for the TCA_CLS_FLAGS_SKIP_SW flag.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
---
 net/sched/cls_bpf.c | 34 +++++++++++++++++++++++++---------
 1 file changed, 25 insertions(+), 9 deletions(-)

diff --git a/net/sched/cls_bpf.c b/net/sched/cls_bpf.c
index 46af423a8a8f..18f9869cd4da 100644
--- a/net/sched/cls_bpf.c
+++ b/net/sched/cls_bpf.c
@@ -28,7 +28,7 @@ MODULE_DESCRIPTION("TC BPF based classifier");
 
 #define CLS_BPF_NAME_LEN	256
 #define CLS_BPF_SUPPORTED_GEN_FLAGS		\
-	TCA_CLS_FLAGS_SKIP_HW
+	(TCA_CLS_FLAGS_SKIP_HW | TCA_CLS_FLAGS_SKIP_SW)
 
 struct cls_bpf_head {
 	struct list_head plist;
@@ -95,7 +95,9 @@ static int cls_bpf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 
 		qdisc_skb_cb(skb)->tc_classid = prog->res.classid;
 
-		if (at_ingress) {
+		if (tc_skip_sw(prog->gen_flags)) {
+			filter_res = prog->exts_integrated ? TC_ACT_UNSPEC : 0;
+		} else if (at_ingress) {
 			/* It is safe to push/pull even if skb_shared() */
 			__skb_push(skb, skb->mac_len);
 			bpf_compute_data_end(skb);
@@ -163,32 +165,42 @@ static int cls_bpf_offload_cmd(struct tcf_proto *tp, struct cls_bpf_prog *prog,
 					     tp->protocol, &offload);
 }
 
-static void cls_bpf_offload(struct tcf_proto *tp, struct cls_bpf_prog *prog,
-			    struct cls_bpf_prog *oldprog)
+static int cls_bpf_offload(struct tcf_proto *tp, struct cls_bpf_prog *prog,
+			   struct cls_bpf_prog *oldprog)
 {
 	struct net_device *dev = tp->q->dev_queue->dev;
 	struct cls_bpf_prog *obj = prog;
 	enum tc_clsbpf_command cmd;
+	bool skip_sw;
+	int ret;
+
+	skip_sw = tc_skip_sw(prog->gen_flags) ||
+		(oldprog && tc_skip_sw(oldprog->gen_flags));
 
 	if (oldprog && oldprog->offloaded) {
 		if (tc_should_offload(dev, tp, prog->gen_flags)) {
 			cmd = TC_CLSBPF_REPLACE;
-		} else {
+		} else if (!tc_skip_sw(prog->gen_flags)) {
 			obj = oldprog;
 			cmd = TC_CLSBPF_DESTROY;
+		} else {
+			return -EINVAL;
 		}
 	} else {
 		if (!tc_should_offload(dev, tp, prog->gen_flags))
-			return;
+			return skip_sw ? -EINVAL : 0;
 		cmd = TC_CLSBPF_ADD;
 	}
 
-	if (cls_bpf_offload_cmd(tp, obj, cmd))
-		return;
+	ret = cls_bpf_offload_cmd(tp, obj, cmd);
+	if (ret)
+		return skip_sw ? ret : 0;
 
 	obj->offloaded = true;
 	if (oldprog)
 		oldprog->offloaded = false;
+
+	return 0;
 }
 
 static void cls_bpf_stop_offload(struct tcf_proto *tp,
@@ -496,7 +508,11 @@ static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
 	if (ret < 0)
 		goto errout;
 
-	cls_bpf_offload(tp, prog, oldprog);
+	ret = cls_bpf_offload(tp, prog, oldprog);
+	if (ret) {
+		cls_bpf_delete_prog(tp, prog);
+		return ret;
+	}
 
 	if (oldprog) {
 		list_replace_rcu(&oldprog->link, &prog->link);
-- 
1.9.1

[PATCHv5 net-next 04/15] bpf: don't (ab)use instructions to store state

[Jakub Kicinski]

Storing state in reserved fields of instructions makes
it impossible to run verifier on programs already
marked as read-only. Allocate and use an array of
per-instruction state instead.

While touching the error path rename and move existing
jump target.

Suggested-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
---
v3:
 - new patch.
---
 kernel/bpf/verifier.c | 51 ++++++++++++++++++++++++++++++++-------------------
 1 file changed, 32 insertions(+), 19 deletions(-)

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 086b3979380c..ce9c0d1721c6 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -181,6 +181,10 @@ struct verifier_stack_elem {
 	struct verifier_stack_elem *next;
 };
 
+struct bpf_insn_aux_data {
+	enum bpf_reg_type ptr_type;	/* pointer type for load/store insns */
+};
+
 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
 
 /* single container for all structs
@@ -196,6 +200,7 @@ struct verifier_env {
 	u32 used_map_cnt;		/* number of used maps */
 	u32 id_gen;			/* used to generate unique reg IDs */
 	bool allow_ptr_leaks;
+	struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */
 };
 
 #define BPF_COMPLEXITY_LIMIT_INSNS	65536
@@ -2340,7 +2345,7 @@ static int do_check(struct verifier_env *env)
 				return err;
 
 		} else if (class == BPF_LDX) {
-			enum bpf_reg_type src_reg_type;
+			enum bpf_reg_type *prev_src_type, src_reg_type;
 
 			/* check for reserved fields is already done */
 
@@ -2370,16 +2375,18 @@ static int do_check(struct verifier_env *env)
 				continue;
 			}
 
-			if (insn->imm == 0) {
+			prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
+
+			if (*prev_src_type == NOT_INIT) {
 				/* saw a valid insn
 				 * dst_reg = *(u32 *)(src_reg + off)
-				 * use reserved 'imm' field to mark this insn
+				 * save type to validate intersecting paths
 				 */
-				insn->imm = src_reg_type;
+				*prev_src_type = src_reg_type;
 
-			} else if (src_reg_type != insn->imm &&
+			} else if (src_reg_type != *prev_src_type &&
 				   (src_reg_type == PTR_TO_CTX ||
-				    insn->imm == PTR_TO_CTX)) {
+				    *prev_src_type == PTR_TO_CTX)) {
 				/* ABuser program is trying to use the same insn
 				 * dst_reg = *(u32*) (src_reg + off)
 				 * with different pointer types:
@@ -2392,7 +2399,7 @@ static int do_check(struct verifier_env *env)
 			}
 
 		} else if (class == BPF_STX) {
-			enum bpf_reg_type dst_reg_type;
+			enum bpf_reg_type *prev_dst_type, dst_reg_type;
 
 			if (BPF_MODE(insn->code) == BPF_XADD) {
 				err = check_xadd(env, insn);
@@ -2420,11 +2427,13 @@ static int do_check(struct verifier_env *env)
 			if (err)
 				return err;
 
-			if (insn->imm == 0) {
-				insn->imm = dst_reg_type;
-			} else if (dst_reg_type != insn->imm &&
+			prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
+
+			if (*prev_dst_type == NOT_INIT) {
+				*prev_dst_type = dst_reg_type;
+			} else if (dst_reg_type != *prev_dst_type &&
 				   (dst_reg_type == PTR_TO_CTX ||
-				    insn->imm == PTR_TO_CTX)) {
+				    *prev_dst_type == PTR_TO_CTX)) {
 				verbose("same insn cannot be used with different pointers\n");
 				return -EINVAL;
 			}
@@ -2703,11 +2712,8 @@ static int convert_ctx_accesses(struct verifier_env *env)
 		else
 			continue;
 
-		if (insn->imm != PTR_TO_CTX) {
-			/* clear internal mark */
-			insn->imm = 0;
+		if (env->insn_aux_data[i].ptr_type != PTR_TO_CTX)
 			continue;
-		}
 
 		cnt = env->prog->aux->ops->
 			convert_ctx_access(type, insn->dst_reg, insn->src_reg,
@@ -2772,6 +2778,11 @@ int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
 	if (!env)
 		return -ENOMEM;
 
+	env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
+				     (*prog)->len);
+	ret = -ENOMEM;
+	if (!env->insn_aux_data)
+		goto err_free_env;
 	env->prog = *prog;
 
 	/* grab the mutex to protect few globals used by verifier */
@@ -2790,12 +2801,12 @@ int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
 		/* log_* values have to be sane */
 		if (log_size < 128 || log_size > UINT_MAX >> 8 ||
 		    log_level == 0 || log_ubuf == NULL)
-			goto free_env;
+			goto err_unlock;
 
 		ret = -ENOMEM;
 		log_buf = vmalloc(log_size);
 		if (!log_buf)
-			goto free_env;
+			goto err_unlock;
 	} else {
 		log_level = 0;
 	}
@@ -2864,14 +2875,16 @@ int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
 free_log_buf:
 	if (log_level)
 		vfree(log_buf);
-free_env:
 	if (!env->prog->aux->used_maps)
 		/* if we didn't copy map pointers into bpf_prog_info, release
 		 * them now. Otherwise free_bpf_prog_info() will release them.
 		 */
 		release_maps(env);
 	*prog = env->prog;
-	kfree(env);
+err_unlock:
 	mutex_unlock(&bpf_verifier_lock);
+	vfree(env->insn_aux_data);
+err_free_env:
+	kfree(env);
 	return ret;
 }
-- 
1.9.1

Re: [PATCHv5 net-next 04/15] bpf: don't (ab)use instructions to store state

[Daniel Borkmann]

On 09/16/2016 11:36 AM, Jakub Kicinski wrote:
> Storing state in reserved fields of instructions makes
> it impossible to run verifier on programs already
> marked as read-only. Allocate and use an array of
> per-instruction state instead.
>
> While touching the error path rename and move existing
> jump target.
>
> Suggested-by: Alexei Starovoitov <ast@kernel.org>
> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> Acked-by: Alexei Starovoitov <ast@kernel.org>

LGMT

Acked-by: Daniel Borkmann <daniel@iogearbox.net>

[PATCHv5 net-next 05/15] bpf: expose internal verfier structures

[Jakub Kicinski]

Move verifier's internal structures to a header file and
prefix their names with bpf_ to avoid potential namespace
conflicts.  Those structures will soon be used by external
analyzers.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
---
v5:
 - fix name of guard defines.
v4:
 - separate from adding the analyzer;
 - squash with the prefixing patch.
---
 include/linux/bpf_verifier.h |  78 +++++++++++++
 kernel/bpf/verifier.c        | 263 +++++++++++++++++--------------------------
 2 files changed, 180 insertions(+), 161 deletions(-)
 create mode 100644 include/linux/bpf_verifier.h

diff --git a/include/linux/bpf_verifier.h b/include/linux/bpf_verifier.h
new file mode 100644
index 000000000000..9056117b4a81
--- /dev/null
+++ b/include/linux/bpf_verifier.h
@@ -0,0 +1,78 @@
+/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of version 2 of the GNU General Public
+ * License as published by the Free Software Foundation.
+ */
+#ifndef _LINUX_BPF_VERIFIER_H
+#define _LINUX_BPF_VERIFIER_H 1
+
+#include <linux/bpf.h> /* for enum bpf_reg_type */
+#include <linux/filter.h> /* for MAX_BPF_STACK */
+
+struct bpf_reg_state {
+	enum bpf_reg_type type;
+	union {
+		/* valid when type == CONST_IMM | PTR_TO_STACK | UNKNOWN_VALUE */
+		s64 imm;
+
+		/* valid when type == PTR_TO_PACKET* */
+		struct {
+			u32 id;
+			u16 off;
+			u16 range;
+		};
+
+		/* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
+		 *   PTR_TO_MAP_VALUE_OR_NULL
+		 */
+		struct bpf_map *map_ptr;
+	};
+};
+
+enum bpf_stack_slot_type {
+	STACK_INVALID,    /* nothing was stored in this stack slot */
+	STACK_SPILL,      /* register spilled into stack */
+	STACK_MISC	  /* BPF program wrote some data into this slot */
+};
+
+#define BPF_REG_SIZE 8	/* size of eBPF register in bytes */
+
+/* state of the program:
+ * type of all registers and stack info
+ */
+struct bpf_verifier_state {
+	struct bpf_reg_state regs[MAX_BPF_REG];
+	u8 stack_slot_type[MAX_BPF_STACK];
+	struct bpf_reg_state spilled_regs[MAX_BPF_STACK / BPF_REG_SIZE];
+};
+
+/* linked list of verifier states used to prune search */
+struct bpf_verifier_state_list {
+	struct bpf_verifier_state state;
+	struct bpf_verifier_state_list *next;
+};
+
+struct bpf_insn_aux_data {
+	enum bpf_reg_type ptr_type;	/* pointer type for load/store insns */
+};
+
+#define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
+
+/* single container for all structs
+ * one verifier_env per bpf_check() call
+ */
+struct bpf_verifier_env {
+	struct bpf_prog *prog;		/* eBPF program being verified */
+	struct bpf_verifier_stack_elem *head; /* stack of verifier states to be processed */
+	int stack_size;			/* number of states to be processed */
+	struct bpf_verifier_state cur_state; /* current verifier state */
+	struct bpf_verifier_state_list **explored_states; /* search pruning optimization */
+	struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
+	u32 used_map_cnt;		/* number of used maps */
+	u32 id_gen;			/* used to generate unique reg IDs */
+	bool allow_ptr_leaks;
+	struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */
+};
+
+#endif /* _LINUX_BPF_VERIFIER_H */
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index ce9c0d1721c6..6e126a417290 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -14,6 +14,7 @@
 #include <linux/types.h>
 #include <linux/slab.h>
 #include <linux/bpf.h>
+#include <linux/bpf_verifier.h>
 #include <linux/filter.h>
 #include <net/netlink.h>
 #include <linux/file.h>
@@ -126,81 +127,16 @@
  * are set to NOT_INIT to indicate that they are no longer readable.
  */
 
-struct reg_state {
-	enum bpf_reg_type type;
-	union {
-		/* valid when type == CONST_IMM | PTR_TO_STACK | UNKNOWN_VALUE */
-		s64 imm;
-
-		/* valid when type == PTR_TO_PACKET* */
-		struct {
-			u32 id;
-			u16 off;
-			u16 range;
-		};
-
-		/* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
-		 *   PTR_TO_MAP_VALUE_OR_NULL
-		 */
-		struct bpf_map *map_ptr;
-	};
-};
-
-enum bpf_stack_slot_type {
-	STACK_INVALID,    /* nothing was stored in this stack slot */
-	STACK_SPILL,      /* register spilled into stack */
-	STACK_MISC	  /* BPF program wrote some data into this slot */
-};
-
-#define BPF_REG_SIZE 8	/* size of eBPF register in bytes */
-
-/* state of the program:
- * type of all registers and stack info
- */
-struct verifier_state {
-	struct reg_state regs[MAX_BPF_REG];
-	u8 stack_slot_type[MAX_BPF_STACK];
-	struct reg_state spilled_regs[MAX_BPF_STACK / BPF_REG_SIZE];
-};
-
-/* linked list of verifier states used to prune search */
-struct verifier_state_list {
-	struct verifier_state state;
-	struct verifier_state_list *next;
-};
-
 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
-struct verifier_stack_elem {
+struct bpf_verifier_stack_elem {
 	/* verifer state is 'st'
 	 * before processing instruction 'insn_idx'
 	 * and after processing instruction 'prev_insn_idx'
 	 */
-	struct verifier_state st;
+	struct bpf_verifier_state st;
 	int insn_idx;
 	int prev_insn_idx;
-	struct verifier_stack_elem *next;
-};
-
-struct bpf_insn_aux_data {
-	enum bpf_reg_type ptr_type;	/* pointer type for load/store insns */
-};
-
-#define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
-
-/* single container for all structs
- * one verifier_env per bpf_check() call
- */
-struct verifier_env {
-	struct bpf_prog *prog;		/* eBPF program being verified */
-	struct verifier_stack_elem *head; /* stack of verifier states to be processed */
-	int stack_size;			/* number of states to be processed */
-	struct verifier_state cur_state; /* current verifier state */
-	struct verifier_state_list **explored_states; /* search pruning optimization */
-	struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
-	u32 used_map_cnt;		/* number of used maps */
-	u32 id_gen;			/* used to generate unique reg IDs */
-	bool allow_ptr_leaks;
-	struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */
+	struct bpf_verifier_stack_elem *next;
 };
 
 #define BPF_COMPLEXITY_LIMIT_INSNS	65536
@@ -252,9 +188,9 @@ static const char * const reg_type_str[] = {
 	[PTR_TO_PACKET_END]	= "pkt_end",
 };
 
-static void print_verifier_state(struct verifier_state *state)
+static void print_verifier_state(struct bpf_verifier_state *state)
 {
-	struct reg_state *reg;
+	struct bpf_reg_state *reg;
 	enum bpf_reg_type t;
 	int i;
 
@@ -430,9 +366,9 @@ static void print_bpf_insn(struct bpf_insn *insn)
 	}
 }
 
-static int pop_stack(struct verifier_env *env, int *prev_insn_idx)
+static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx)
 {
-	struct verifier_stack_elem *elem;
+	struct bpf_verifier_stack_elem *elem;
 	int insn_idx;
 
 	if (env->head == NULL)
@@ -449,12 +385,12 @@ static int pop_stack(struct verifier_env *env, int *prev_insn_idx)
 	return insn_idx;
 }
 
-static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx,
-					 int prev_insn_idx)
+static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env,
+					     int insn_idx, int prev_insn_idx)
 {
-	struct verifier_stack_elem *elem;
+	struct bpf_verifier_stack_elem *elem;
 
-	elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL);
+	elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
 	if (!elem)
 		goto err;
 
@@ -480,7 +416,7 @@ static const int caller_saved[CALLER_SAVED_REGS] = {
 	BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
 };
 
-static void init_reg_state(struct reg_state *regs)
+static void init_reg_state(struct bpf_reg_state *regs)
 {
 	int i;
 
@@ -496,7 +432,7 @@ static void init_reg_state(struct reg_state *regs)
 	regs[BPF_REG_1].type = PTR_TO_CTX;
 }
 
-static void mark_reg_unknown_value(struct reg_state *regs, u32 regno)
+static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
 {
 	BUG_ON(regno >= MAX_BPF_REG);
 	regs[regno].type = UNKNOWN_VALUE;
@@ -509,7 +445,7 @@ enum reg_arg_type {
 	DST_OP_NO_MARK	/* same as above, check only, don't mark */
 };
 
-static int check_reg_arg(struct reg_state *regs, u32 regno,
+static int check_reg_arg(struct bpf_reg_state *regs, u32 regno,
 			 enum reg_arg_type t)
 {
 	if (regno >= MAX_BPF_REG) {
@@ -569,8 +505,8 @@ static bool is_spillable_regtype(enum bpf_reg_type type)
 /* check_stack_read/write functions track spill/fill of registers,
  * stack boundary and alignment are checked in check_mem_access()
  */
-static int check_stack_write(struct verifier_state *state, int off, int size,
-			     int value_regno)
+static int check_stack_write(struct bpf_verifier_state *state, int off,
+			     int size, int value_regno)
 {
 	int i;
 	/* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
@@ -595,7 +531,7 @@ static int check_stack_write(struct verifier_state *state, int off, int size,
 	} else {
 		/* regular write of data into stack */
 		state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
-			(struct reg_state) {};
+			(struct bpf_reg_state) {};
 
 		for (i = 0; i < size; i++)
 			state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
@@ -603,7 +539,7 @@ static int check_stack_write(struct verifier_state *state, int off, int size,
 	return 0;
 }
 
-static int check_stack_read(struct verifier_state *state, int off, int size,
+static int check_stack_read(struct bpf_verifier_state *state, int off, int size,
 			    int value_regno)
 {
 	u8 *slot_type;
@@ -644,7 +580,7 @@ static int check_stack_read(struct verifier_state *state, int off, int size,
 }
 
 /* check read/write into map element returned by bpf_map_lookup_elem() */
-static int check_map_access(struct verifier_env *env, u32 regno, int off,
+static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off,
 			    int size)
 {
 	struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
@@ -669,11 +605,11 @@ static bool may_write_pkt_data(enum bpf_prog_type type)
 	}
 }
 
-static int check_packet_access(struct verifier_env *env, u32 regno, int off,
+static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
 			       int size)
 {
-	struct reg_state *regs = env->cur_state.regs;
-	struct reg_state *reg = &regs[regno];
+	struct bpf_reg_state *regs = env->cur_state.regs;
+	struct bpf_reg_state *reg = &regs[regno];
 
 	off += reg->off;
 	if (off < 0 || off + size > reg->range) {
@@ -685,7 +621,7 @@ static int check_packet_access(struct verifier_env *env, u32 regno, int off,
 }
 
 /* check access to 'struct bpf_context' fields */
-static int check_ctx_access(struct verifier_env *env, int off, int size,
+static int check_ctx_access(struct bpf_verifier_env *env, int off, int size,
 			    enum bpf_access_type t, enum bpf_reg_type *reg_type)
 {
 	if (env->prog->aux->ops->is_valid_access &&
@@ -700,7 +636,7 @@ static int check_ctx_access(struct verifier_env *env, int off, int size,
 	return -EACCES;
 }
 
-static bool is_pointer_value(struct verifier_env *env, int regno)
+static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
 {
 	if (env->allow_ptr_leaks)
 		return false;
@@ -714,12 +650,13 @@ static bool is_pointer_value(struct verifier_env *env, int regno)
 	}
 }
 
-static int check_ptr_alignment(struct verifier_env *env, struct reg_state *reg,
-			       int off, int size)
+static int check_ptr_alignment(struct bpf_verifier_env *env,
+			       struct bpf_reg_state *reg, int off, int size)
 {
 	if (reg->type != PTR_TO_PACKET) {
 		if (off % size != 0) {
-			verbose("misaligned access off %d size %d\n", off, size);
+			verbose("misaligned access off %d size %d\n",
+				off, size);
 			return -EACCES;
 		} else {
 			return 0;
@@ -760,12 +697,12 @@ static int check_ptr_alignment(struct verifier_env *env, struct reg_state *reg,
  * if t==write && value_regno==-1, some unknown value is stored into memory
  * if t==read && value_regno==-1, don't care what we read from memory
  */
-static int check_mem_access(struct verifier_env *env, u32 regno, int off,
+static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off,
 			    int bpf_size, enum bpf_access_type t,
 			    int value_regno)
 {
-	struct verifier_state *state = &env->cur_state;
-	struct reg_state *reg = &state->regs[regno];
+	struct bpf_verifier_state *state = &env->cur_state;
+	struct bpf_reg_state *reg = &state->regs[regno];
 	int size, err = 0;
 
 	if (reg->type == PTR_TO_STACK)
@@ -851,9 +788,9 @@ static int check_mem_access(struct verifier_env *env, u32 regno, int off,
 	return err;
 }
 
-static int check_xadd(struct verifier_env *env, struct bpf_insn *insn)
+static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn)
 {
-	struct reg_state *regs = env->cur_state.regs;
+	struct bpf_reg_state *regs = env->cur_state.regs;
 	int err;
 
 	if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
@@ -887,12 +824,12 @@ static int check_xadd(struct verifier_env *env, struct bpf_insn *insn)
  * bytes from that pointer, make sure that it's within stack boundary
  * and all elements of stack are initialized
  */
-static int check_stack_boundary(struct verifier_env *env, int regno,
+static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
 				int access_size, bool zero_size_allowed,
 				struct bpf_call_arg_meta *meta)
 {
-	struct verifier_state *state = &env->cur_state;
-	struct reg_state *regs = state->regs;
+	struct bpf_verifier_state *state = &env->cur_state;
+	struct bpf_reg_state *regs = state->regs;
 	int off, i;
 
 	if (regs[regno].type != PTR_TO_STACK) {
@@ -931,11 +868,11 @@ static int check_stack_boundary(struct verifier_env *env, int regno,
 	return 0;
 }
 
-static int check_func_arg(struct verifier_env *env, u32 regno,
+static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
 			  enum bpf_arg_type arg_type,
 			  struct bpf_call_arg_meta *meta)
 {
-	struct reg_state *regs = env->cur_state.regs, *reg = &regs[regno];
+	struct bpf_reg_state *regs = env->cur_state.regs, *reg = &regs[regno];
 	enum bpf_reg_type expected_type, type = reg->type;
 	int err = 0;
 
@@ -1140,10 +1077,10 @@ static int check_raw_mode(const struct bpf_func_proto *fn)
 	return count > 1 ? -EINVAL : 0;
 }
 
-static void clear_all_pkt_pointers(struct verifier_env *env)
+static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
 {
-	struct verifier_state *state = &env->cur_state;
-	struct reg_state *regs = state->regs, *reg;
+	struct bpf_verifier_state *state = &env->cur_state;
+	struct bpf_reg_state *regs = state->regs, *reg;
 	int i;
 
 	for (i = 0; i < MAX_BPF_REG; i++)
@@ -1163,12 +1100,12 @@ static void clear_all_pkt_pointers(struct verifier_env *env)
 	}
 }
 
-static int check_call(struct verifier_env *env, int func_id)
+static int check_call(struct bpf_verifier_env *env, int func_id)
 {
-	struct verifier_state *state = &env->cur_state;
+	struct bpf_verifier_state *state = &env->cur_state;
 	const struct bpf_func_proto *fn = NULL;
-	struct reg_state *regs = state->regs;
-	struct reg_state *reg;
+	struct bpf_reg_state *regs = state->regs;
+	struct bpf_reg_state *reg;
 	struct bpf_call_arg_meta meta;
 	bool changes_data;
 	int i, err;
@@ -1270,12 +1207,13 @@ static int check_call(struct verifier_env *env, int func_id)
 	return 0;
 }
 
-static int check_packet_ptr_add(struct verifier_env *env, struct bpf_insn *insn)
+static int check_packet_ptr_add(struct bpf_verifier_env *env,
+				struct bpf_insn *insn)
 {
-	struct reg_state *regs = env->cur_state.regs;
-	struct reg_state *dst_reg = &regs[insn->dst_reg];
-	struct reg_state *src_reg = &regs[insn->src_reg];
-	struct reg_state tmp_reg;
+	struct bpf_reg_state *regs = env->cur_state.regs;
+	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
+	struct bpf_reg_state *src_reg = &regs[insn->src_reg];
+	struct bpf_reg_state tmp_reg;
 	s32 imm;
 
 	if (BPF_SRC(insn->code) == BPF_K) {
@@ -1343,10 +1281,10 @@ static int check_packet_ptr_add(struct verifier_env *env, struct bpf_insn *insn)
 	return 0;
 }
 
-static int evaluate_reg_alu(struct verifier_env *env, struct bpf_insn *insn)
+static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn)
 {
-	struct reg_state *regs = env->cur_state.regs;
-	struct reg_state *dst_reg = &regs[insn->dst_reg];
+	struct bpf_reg_state *regs = env->cur_state.regs;
+	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
 	u8 opcode = BPF_OP(insn->code);
 	s64 imm_log2;
 
@@ -1356,7 +1294,7 @@ static int evaluate_reg_alu(struct verifier_env *env, struct bpf_insn *insn)
 	 */
 
 	if (BPF_SRC(insn->code) == BPF_X) {
-		struct reg_state *src_reg = &regs[insn->src_reg];
+		struct bpf_reg_state *src_reg = &regs[insn->src_reg];
 
 		if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
 		    dst_reg->imm && opcode == BPF_ADD) {
@@ -1445,11 +1383,12 @@ static int evaluate_reg_alu(struct verifier_env *env, struct bpf_insn *insn)
 	return 0;
 }
 
-static int evaluate_reg_imm_alu(struct verifier_env *env, struct bpf_insn *insn)
+static int evaluate_reg_imm_alu(struct bpf_verifier_env *env,
+				struct bpf_insn *insn)
 {
-	struct reg_state *regs = env->cur_state.regs;
-	struct reg_state *dst_reg = &regs[insn->dst_reg];
-	struct reg_state *src_reg = &regs[insn->src_reg];
+	struct bpf_reg_state *regs = env->cur_state.regs;
+	struct bpf_reg_state *dst_reg = &regs[insn->dst_reg];
+	struct bpf_reg_state *src_reg = &regs[insn->src_reg];
 	u8 opcode = BPF_OP(insn->code);
 
 	/* dst_reg->type == CONST_IMM here, simulate execution of 'add' insn.
@@ -1466,9 +1405,9 @@ static int evaluate_reg_imm_alu(struct verifier_env *env, struct bpf_insn *insn)
 }
 
 /* check validity of 32-bit and 64-bit arithmetic operations */
-static int check_alu_op(struct verifier_env *env, struct bpf_insn *insn)
+static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
 {
-	struct reg_state *regs = env->cur_state.regs, *dst_reg;
+	struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
 	u8 opcode = BPF_OP(insn->code);
 	int err;
 
@@ -1648,10 +1587,10 @@ static int check_alu_op(struct verifier_env *env, struct bpf_insn *insn)
 	return 0;
 }
 
-static void find_good_pkt_pointers(struct verifier_state *state,
-				   const struct reg_state *dst_reg)
+static void find_good_pkt_pointers(struct bpf_verifier_state *state,
+				   struct bpf_reg_state *dst_reg)
 {
-	struct reg_state *regs = state->regs, *reg;
+	struct bpf_reg_state *regs = state->regs, *reg;
 	int i;
 
 	/* LLVM can generate two kind of checks:
@@ -1697,11 +1636,11 @@ static void find_good_pkt_pointers(struct verifier_state *state,
 	}
 }
 
-static int check_cond_jmp_op(struct verifier_env *env,
+static int check_cond_jmp_op(struct bpf_verifier_env *env,
 			     struct bpf_insn *insn, int *insn_idx)
 {
-	struct verifier_state *other_branch, *this_branch = &env->cur_state;
-	struct reg_state *regs = this_branch->regs, *dst_reg;
+	struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state;
+	struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
 	u8 opcode = BPF_OP(insn->code);
 	int err;
 
@@ -1763,7 +1702,7 @@ static int check_cond_jmp_op(struct verifier_env *env,
 	if (!other_branch)
 		return -EFAULT;
 
-	/* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
+	/* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
 	if (BPF_SRC(insn->code) == BPF_K &&
 	    insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
 	    dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
@@ -1805,9 +1744,9 @@ static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
 }
 
 /* verify BPF_LD_IMM64 instruction */
-static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
+static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
 {
-	struct reg_state *regs = env->cur_state.regs;
+	struct bpf_reg_state *regs = env->cur_state.regs;
 	int err;
 
 	if (BPF_SIZE(insn->code) != BPF_DW) {
@@ -1862,11 +1801,11 @@ static bool may_access_skb(enum bpf_prog_type type)
  * Output:
  *   R0 - 8/16/32-bit skb data converted to cpu endianness
  */
-static int check_ld_abs(struct verifier_env *env, struct bpf_insn *insn)
+static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
 {
-	struct reg_state *regs = env->cur_state.regs;
+	struct bpf_reg_state *regs = env->cur_state.regs;
 	u8 mode = BPF_MODE(insn->code);
-	struct reg_state *reg;
+	struct bpf_reg_state *reg;
 	int i, err;
 
 	if (!may_access_skb(env->prog->type)) {
@@ -1952,7 +1891,7 @@ enum {
 	BRANCH = 2,
 };
 
-#define STATE_LIST_MARK ((struct verifier_state_list *) -1L)
+#define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
 
 static int *insn_stack;	/* stack of insns to process */
 static int cur_stack;	/* current stack index */
@@ -1963,7 +1902,7 @@ static int *insn_state;
  * w - next instruction
  * e - edge
  */
-static int push_insn(int t, int w, int e, struct verifier_env *env)
+static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
 {
 	if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
 		return 0;
@@ -2004,7 +1943,7 @@ static int push_insn(int t, int w, int e, struct verifier_env *env)
 /* non-recursive depth-first-search to detect loops in BPF program
  * loop == back-edge in directed graph
  */
-static int check_cfg(struct verifier_env *env)
+static int check_cfg(struct bpf_verifier_env *env)
 {
 	struct bpf_insn *insns = env->prog->insnsi;
 	int insn_cnt = env->prog->len;
@@ -2113,7 +2052,8 @@ static int check_cfg(struct verifier_env *env)
 /* the following conditions reduce the number of explored insns
  * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
  */
-static bool compare_ptrs_to_packet(struct reg_state *old, struct reg_state *cur)
+static bool compare_ptrs_to_packet(struct bpf_reg_state *old,
+				   struct bpf_reg_state *cur)
 {
 	if (old->id != cur->id)
 		return false;
@@ -2188,9 +2128,10 @@ static bool compare_ptrs_to_packet(struct reg_state *old, struct reg_state *cur)
  * whereas register type in current state is meaningful, it means that
  * the current state will reach 'bpf_exit' instruction safely
  */
-static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
+static bool states_equal(struct bpf_verifier_state *old,
+			 struct bpf_verifier_state *cur)
 {
-	struct reg_state *rold, *rcur;
+	struct bpf_reg_state *rold, *rcur;
 	int i;
 
 	for (i = 0; i < MAX_BPF_REG; i++) {
@@ -2230,9 +2171,9 @@ static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
 			 * the same, check that stored pointers types
 			 * are the same as well.
 			 * Ex: explored safe path could have stored
-			 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8}
+			 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
 			 * but current path has stored:
-			 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16}
+			 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
 			 * such verifier states are not equivalent.
 			 * return false to continue verification of this path
 			 */
@@ -2243,10 +2184,10 @@ static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
 	return true;
 }
 
-static int is_state_visited(struct verifier_env *env, int insn_idx)
+static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
 {
-	struct verifier_state_list *new_sl;
-	struct verifier_state_list *sl;
+	struct bpf_verifier_state_list *new_sl;
+	struct bpf_verifier_state_list *sl;
 
 	sl = env->explored_states[insn_idx];
 	if (!sl)
@@ -2270,7 +2211,7 @@ static int is_state_visited(struct verifier_env *env, int insn_idx)
 	 * it will be rejected. Since there are no loops, we won't be
 	 * seeing this 'insn_idx' instruction again on the way to bpf_exit
 	 */
-	new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER);
+	new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER);
 	if (!new_sl)
 		return -ENOMEM;
 
@@ -2281,11 +2222,11 @@ static int is_state_visited(struct verifier_env *env, int insn_idx)
 	return 0;
 }
 
-static int do_check(struct verifier_env *env)
+static int do_check(struct bpf_verifier_env *env)
 {
-	struct verifier_state *state = &env->cur_state;
+	struct bpf_verifier_state *state = &env->cur_state;
 	struct bpf_insn *insns = env->prog->insnsi;
-	struct reg_state *regs = state->regs;
+	struct bpf_reg_state *regs = state->regs;
 	int insn_cnt = env->prog->len;
 	int insn_idx, prev_insn_idx = 0;
 	int insn_processed = 0;
@@ -2568,7 +2509,7 @@ static int check_map_prog_compatibility(struct bpf_map *map,
 /* look for pseudo eBPF instructions that access map FDs and
  * replace them with actual map pointers
  */
-static int replace_map_fd_with_map_ptr(struct verifier_env *env)
+static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
 {
 	struct bpf_insn *insn = env->prog->insnsi;
 	int insn_cnt = env->prog->len;
@@ -2665,7 +2606,7 @@ static int replace_map_fd_with_map_ptr(struct verifier_env *env)
 }
 
 /* drop refcnt of maps used by the rejected program */
-static void release_maps(struct verifier_env *env)
+static void release_maps(struct bpf_verifier_env *env)
 {
 	int i;
 
@@ -2674,7 +2615,7 @@ static void release_maps(struct verifier_env *env)
 }
 
 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
-static void convert_pseudo_ld_imm64(struct verifier_env *env)
+static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
 {
 	struct bpf_insn *insn = env->prog->insnsi;
 	int insn_cnt = env->prog->len;
@@ -2688,7 +2629,7 @@ static void convert_pseudo_ld_imm64(struct verifier_env *env)
 /* convert load instructions that access fields of 'struct __sk_buff'
  * into sequence of instructions that access fields of 'struct sk_buff'
  */
-static int convert_ctx_accesses(struct verifier_env *env)
+static int convert_ctx_accesses(struct bpf_verifier_env *env)
 {
 	struct bpf_insn *insn = env->prog->insnsi;
 	int insn_cnt = env->prog->len;
@@ -2740,9 +2681,9 @@ static int convert_ctx_accesses(struct verifier_env *env)
 	return 0;
 }
 
-static void free_states(struct verifier_env *env)
+static void free_states(struct bpf_verifier_env *env)
 {
-	struct verifier_state_list *sl, *sln;
+	struct bpf_verifier_state_list *sl, *sln;
 	int i;
 
 	if (!env->explored_states)
@@ -2765,16 +2706,16 @@ static void free_states(struct verifier_env *env)
 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
 {
 	char __user *log_ubuf = NULL;
-	struct verifier_env *env;
+	struct bpf_verifier_env *env;
 	int ret = -EINVAL;
 
 	if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS)
 		return -E2BIG;
 
-	/* 'struct verifier_env' can be global, but since it's not small,
+	/* 'struct bpf_verifier_env' can be global, but since it's not small,
 	 * allocate/free it every time bpf_check() is called
 	 */
-	env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL);
+	env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
 	if (!env)
 		return -ENOMEM;
 
@@ -2816,7 +2757,7 @@ int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
 		goto skip_full_check;
 
 	env->explored_states = kcalloc(env->prog->len,
-				       sizeof(struct verifier_state_list *),
+				       sizeof(struct bpf_verifier_state_list *),
 				       GFP_USER);
 	ret = -ENOMEM;
 	if (!env->explored_states)
-- 
1.9.1

Re: [PATCHv5 net-next 05/15] bpf: expose internal verfier structures

[Daniel Borkmann]

On 09/16/2016 11:36 AM, Jakub Kicinski wrote:
> Move verifier's internal structures to a header file and
> prefix their names with bpf_ to avoid potential namespace
> conflicts.  Those structures will soon be used by external
> analyzers.
>
> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> Acked-by: Alexei Starovoitov <ast@kernel.org>

Acked-by: Daniel Borkmann <daniel@iogearbox.net>

[PATCHv5 net-next 06/15] bpf: enable non-core use of the verfier

[Jakub Kicinski]

Advanced JIT compilers and translators may want to use
eBPF verifier as a base for parsers or to perform custom
checks and validations.

Add ability for external users to invoke the verifier
and provide callbacks to be invoked for every intruction
checked.  For now only add most basic callback for
per-instruction pre-interpretation checks is added.  More
advanced users may also like to have per-instruction post
callback and state comparison callback.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
---
v4:
 - separate from the header split patch.
---
 include/linux/bpf_verifier.h | 11 +++++++
 kernel/bpf/verifier.c        | 68 ++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 79 insertions(+)

diff --git a/include/linux/bpf_verifier.h b/include/linux/bpf_verifier.h
index 9056117b4a81..925359e1d9a1 100644
--- a/include/linux/bpf_verifier.h
+++ b/include/linux/bpf_verifier.h
@@ -59,6 +59,12 @@ struct bpf_insn_aux_data {
 
 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
 
+struct bpf_verifier_env;
+struct bpf_ext_analyzer_ops {
+	int (*insn_hook)(struct bpf_verifier_env *env,
+			 int insn_idx, int prev_insn_idx);
+};
+
 /* single container for all structs
  * one verifier_env per bpf_check() call
  */
@@ -68,6 +74,8 @@ struct bpf_verifier_env {
 	int stack_size;			/* number of states to be processed */
 	struct bpf_verifier_state cur_state; /* current verifier state */
 	struct bpf_verifier_state_list **explored_states; /* search pruning optimization */
+	const struct bpf_ext_analyzer_ops *analyzer_ops; /* external analyzer ops */
+	void *analyzer_priv; /* pointer to external analyzer's private data */
 	struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
 	u32 used_map_cnt;		/* number of used maps */
 	u32 id_gen;			/* used to generate unique reg IDs */
@@ -75,4 +83,7 @@ struct bpf_verifier_env {
 	struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */
 };
 
+int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
+		 void *priv);
+
 #endif /* _LINUX_BPF_VERIFIER_H */
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 6e126a417290..d93e78331b90 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -624,6 +624,10 @@ static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
 static int check_ctx_access(struct bpf_verifier_env *env, int off, int size,
 			    enum bpf_access_type t, enum bpf_reg_type *reg_type)
 {
+	/* for analyzer ctx accesses are already validated and converted */
+	if (env->analyzer_ops)
+		return 0;
+
 	if (env->prog->aux->ops->is_valid_access &&
 	    env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) {
 		/* remember the offset of last byte accessed in ctx */
@@ -2222,6 +2226,15 @@ static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
 	return 0;
 }
 
+static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
+				  int insn_idx, int prev_insn_idx)
+{
+	if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
+		return 0;
+
+	return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
+}
+
 static int do_check(struct bpf_verifier_env *env)
 {
 	struct bpf_verifier_state *state = &env->cur_state;
@@ -2280,6 +2293,10 @@ static int do_check(struct bpf_verifier_env *env)
 			print_bpf_insn(insn);
 		}
 
+		err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
+		if (err)
+			return err;
+
 		if (class == BPF_ALU || class == BPF_ALU64) {
 			err = check_alu_op(env, insn);
 			if (err)
@@ -2829,3 +2846,54 @@ int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
 	kfree(env);
 	return ret;
 }
+
+int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
+		 void *priv)
+{
+	struct bpf_verifier_env *env;
+	int ret;
+
+	env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
+	if (!env)
+		return -ENOMEM;
+
+	env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
+				     prog->len);
+	ret = -ENOMEM;
+	if (!env->insn_aux_data)
+		goto err_free_env;
+	env->prog = prog;
+	env->analyzer_ops = ops;
+	env->analyzer_priv = priv;
+
+	/* grab the mutex to protect few globals used by verifier */
+	mutex_lock(&bpf_verifier_lock);
+
+	log_level = 0;
+
+	env->explored_states = kcalloc(env->prog->len,
+				       sizeof(struct bpf_verifier_state_list *),
+				       GFP_KERNEL);
+	ret = -ENOMEM;
+	if (!env->explored_states)
+		goto skip_full_check;
+
+	ret = check_cfg(env);
+	if (ret < 0)
+		goto skip_full_check;
+
+	env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
+
+	ret = do_check(env);
+
+skip_full_check:
+	while (pop_stack(env, NULL) >= 0);
+	free_states(env);
+
+	mutex_unlock(&bpf_verifier_lock);
+	vfree(env->insn_aux_data);
+err_free_env:
+	kfree(env);
+	return ret;
+}
+EXPORT_SYMBOL_GPL(bpf_analyzer);
-- 
1.9.1

Re: [PATCHv5 net-next 06/15] bpf: enable non-core use of the verfier

[Daniel Borkmann]

On 09/16/2016 11:36 AM, Jakub Kicinski wrote:
> Advanced JIT compilers and translators may want to use
> eBPF verifier as a base for parsers or to perform custom
> checks and validations.
>
> Add ability for external users to invoke the verifier
> and provide callbacks to be invoked for every intruction
> checked.  For now only add most basic callback for
> per-instruction pre-interpretation checks is added.  More
> advanced users may also like to have per-instruction post
> callback and state comparison callback.
>
> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> Acked-by: Alexei Starovoitov <ast@kernel.org>
> ---
> v4:
>   - separate from the header split patch.
> ---
>   include/linux/bpf_verifier.h | 11 +++++++
>   kernel/bpf/verifier.c        | 68 ++++++++++++++++++++++++++++++++++++++++++++
>   2 files changed, 79 insertions(+)
>
> diff --git a/include/linux/bpf_verifier.h b/include/linux/bpf_verifier.h
> index 9056117b4a81..925359e1d9a1 100644
> --- a/include/linux/bpf_verifier.h
> +++ b/include/linux/bpf_verifier.h
> @@ -59,6 +59,12 @@ struct bpf_insn_aux_data {
>
>   #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
>
> +struct bpf_verifier_env;
> +struct bpf_ext_analyzer_ops {
> +	int (*insn_hook)(struct bpf_verifier_env *env,
> +			 int insn_idx, int prev_insn_idx);
> +};
> +
>   /* single container for all structs
>    * one verifier_env per bpf_check() call
>    */
> @@ -68,6 +74,8 @@ struct bpf_verifier_env {
>   	int stack_size;			/* number of states to be processed */
>   	struct bpf_verifier_state cur_state; /* current verifier state */
>   	struct bpf_verifier_state_list **explored_states; /* search pruning optimization */
> +	const struct bpf_ext_analyzer_ops *analyzer_ops; /* external analyzer ops */
> +	void *analyzer_priv; /* pointer to external analyzer's private data */
>   	struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
>   	u32 used_map_cnt;		/* number of used maps */
>   	u32 id_gen;			/* used to generate unique reg IDs */
> @@ -75,4 +83,7 @@ struct bpf_verifier_env {
>   	struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */
>   };
>
> +int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
> +		 void *priv);
> +
>   #endif /* _LINUX_BPF_VERIFIER_H */
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index 6e126a417290..d93e78331b90 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -624,6 +624,10 @@ static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
>   static int check_ctx_access(struct bpf_verifier_env *env, int off, int size,
>   			    enum bpf_access_type t, enum bpf_reg_type *reg_type)
>   {
> +	/* for analyzer ctx accesses are already validated and converted */
> +	if (env->analyzer_ops)
> +		return 0;
> +
>   	if (env->prog->aux->ops->is_valid_access &&
>   	    env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) {
>   		/* remember the offset of last byte accessed in ctx */
> @@ -2222,6 +2226,15 @@ static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
>   	return 0;
>   }
>
> +static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
> +				  int insn_idx, int prev_insn_idx)
> +{
> +	if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
> +		return 0;
> +
> +	return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
> +}
> +
>   static int do_check(struct bpf_verifier_env *env)
>   {
>   	struct bpf_verifier_state *state = &env->cur_state;
> @@ -2280,6 +2293,10 @@ static int do_check(struct bpf_verifier_env *env)
>   			print_bpf_insn(insn);
>   		}
>
> +		err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
> +		if (err)
> +			return err;
> +

Looking at this and the nfp code translator (patch 8/15): Did you check this
also with JIT hardening enabled? Presumably nfp sees this after it got JITed
through the normal load via syscall. Then, when this gets rewritten using the
BPF_REG_AX helper before the image gets locked, and you later on push this
through bpf_analyzer() again, where the hook is called before re-verification
of insns, it's still assumed to be MAX_BPF_REG in your hook callbacks, right?
So, I was wondering wrt out of bounds in nfp_verify_insn() -> nfp_bpf_check_ctx_ptr()
for things like BPF_STX? If that's the case, it would make sense to just reject
any prog with reg that is BPF_REG_AX in nfp_verify_insn() upfront. Alternative
would be to use MAX_BPF_JIT_REG in nfp and let bpf_analyzer() fail this via
check_reg_arg() test.

>   		if (class == BPF_ALU || class == BPF_ALU64) {
>   			err = check_alu_op(env, insn);
>   			if (err)
> @@ -2829,3 +2846,54 @@ int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
>   	kfree(env);
>   	return ret;
>   }
> +
> +int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
> +		 void *priv)
> +{
> +	struct bpf_verifier_env *env;
> +	int ret;
> +
> +	env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
> +	if (!env)
> +		return -ENOMEM;
> +
> +	env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
> +				     prog->len);
> +	ret = -ENOMEM;
> +	if (!env->insn_aux_data)
> +		goto err_free_env;
> +	env->prog = prog;
> +	env->analyzer_ops = ops;
> +	env->analyzer_priv = priv;
> +
> +	/* grab the mutex to protect few globals used by verifier */
> +	mutex_lock(&bpf_verifier_lock);
> +
> +	log_level = 0;
> +
> +	env->explored_states = kcalloc(env->prog->len,
> +				       sizeof(struct bpf_verifier_state_list *),
> +				       GFP_KERNEL);
> +	ret = -ENOMEM;
> +	if (!env->explored_states)
> +		goto skip_full_check;
> +
> +	ret = check_cfg(env);
> +	if (ret < 0)
> +		goto skip_full_check;
> +
> +	env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
> +
> +	ret = do_check(env);
> +
> +skip_full_check:
> +	while (pop_stack(env, NULL) >= 0);
> +	free_states(env);
> +
> +	mutex_unlock(&bpf_verifier_lock);
> +	vfree(env->insn_aux_data);
> +err_free_env:
> +	kfree(env);
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(bpf_analyzer);
>

[PATCHv5 net-next 07/15] bpf: recognize 64bit immediate loads as consts

[Jakub Kicinski]

When running as parser interpret BPF_LD | BPF_IMM | BPF_DW
instructions as loading CONST_IMM with the value stored
in imm.  The verifier will continue not recognizing those
due to concerns about search space/program complexity
increase.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
---
v3:
 - limit to parsers.
---
 kernel/bpf/verifier.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index d93e78331b90..f5bed7cce08d 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -1766,9 +1766,19 @@ static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
 	if (err)
 		return err;
 
-	if (insn->src_reg == 0)
-		/* generic move 64-bit immediate into a register */
+	if (insn->src_reg == 0) {
+		/* generic move 64-bit immediate into a register,
+		 * only analyzer needs to collect the ld_imm value.
+		 */
+		u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
+
+		if (!env->analyzer_ops)
+			return 0;
+
+		regs[insn->dst_reg].type = CONST_IMM;
+		regs[insn->dst_reg].imm = imm;
 		return 0;
+	}
 
 	/* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
 	BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
-- 
1.9.1

Re: [PATCHv5 net-next 07/15] bpf: recognize 64bit immediate loads as consts

[Daniel Borkmann]

On 09/16/2016 11:36 AM, Jakub Kicinski wrote:
> When running as parser interpret BPF_LD | BPF_IMM | BPF_DW
> instructions as loading CONST_IMM with the value stored
> in imm.  The verifier will continue not recognizing those
> due to concerns about search space/program complexity
> increase.
>
> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> Acked-by: Alexei Starovoitov <ast@kernel.org>

Acked-by: Daniel Borkmann <daniel@iogearbox.net>

[PATCHv5 net-next 08/15] nfp: add BPF to NFP code translator

[Jakub Kicinski]

Add translator for JITing eBPF to operations which
can be executed on NFP's programmable engines.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
v4:
 - use bitfield.h directly.
v3:
 - don't clone the program for the verifier (no longer needed);
 - temporarily add a local copy of macros from bitfield.h.
---
 drivers/net/ethernet/netronome/nfp/Makefile        |    6 +
 drivers/net/ethernet/netronome/nfp/nfp_asm.h       |  233 +++
 drivers/net/ethernet/netronome/nfp/nfp_bpf.h       |  208 +++
 drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c   | 1729 ++++++++++++++++++++
 .../net/ethernet/netronome/nfp/nfp_bpf_verifier.c  |  151 ++
 5 files changed, 2327 insertions(+)
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_asm.h
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_bpf.h
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_bpf_verifier.c

diff --git a/drivers/net/ethernet/netronome/nfp/Makefile b/drivers/net/ethernet/netronome/nfp/Makefile
index 68178819ff12..5f12689bf523 100644
--- a/drivers/net/ethernet/netronome/nfp/Makefile
+++ b/drivers/net/ethernet/netronome/nfp/Makefile
@@ -5,4 +5,10 @@ nfp_netvf-objs := \
 	    nfp_net_ethtool.o \
 	    nfp_netvf_main.o
 
+ifeq ($(CONFIG_BPF_SYSCALL),y)
+nfp_netvf-objs += \
+	    nfp_bpf_verifier.o \
+	    nfp_bpf_jit.o
+endif
+
 nfp_netvf-$(CONFIG_NFP_NET_DEBUG) += nfp_net_debugfs.o
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_asm.h b/drivers/net/ethernet/netronome/nfp/nfp_asm.h
new file mode 100644
index 000000000000..22484b6fd3e8
--- /dev/null
+++ b/drivers/net/ethernet/netronome/nfp/nfp_asm.h
@@ -0,0 +1,233 @@
+/*
+ * 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.
+ */
+
+#ifndef __NFP_ASM_H__
+#define __NFP_ASM_H__ 1
+
+#include "nfp_bpf.h"
+
+#define REG_NONE	0
+
+#define RE_REG_NO_DST	0x020
+#define RE_REG_IMM	0x020
+#define RE_REG_IMM_encode(x)					\
+	(RE_REG_IMM | ((x) & 0x1f) | (((x) & 0x60) << 1))
+#define RE_REG_IMM_MAX	 0x07fULL
+#define RE_REG_XFR	0x080
+
+#define UR_REG_XFR	0x180
+#define UR_REG_NN	0x280
+#define UR_REG_NO_DST	0x300
+#define UR_REG_IMM	UR_REG_NO_DST
+#define UR_REG_IMM_encode(x) (UR_REG_IMM | (x))
+#define UR_REG_IMM_MAX	 0x0ffULL
+
+#define OP_BR_BASE	0x0d800000020ULL
+#define OP_BR_BASE_MASK	0x0f8000c3ce0ULL
+#define OP_BR_MASK	0x0000000001fULL
+#define OP_BR_EV_PIP	0x00000000300ULL
+#define OP_BR_CSS	0x0000003c000ULL
+#define OP_BR_DEFBR	0x00000300000ULL
+#define OP_BR_ADDR_LO	0x007ffc00000ULL
+#define OP_BR_ADDR_HI	0x10000000000ULL
+
+#define nfp_is_br(_insn)				\
+	(((_insn) & OP_BR_BASE_MASK) == OP_BR_BASE)
+
+enum br_mask {
+	BR_BEQ = 0x00,
+	BR_BNE = 0x01,
+	BR_BHS = 0x04,
+	BR_BLO = 0x05,
+	BR_BGE = 0x08,
+	BR_UNC = 0x18,
+};
+
+enum br_ev_pip {
+	BR_EV_PIP_UNCOND = 0,
+	BR_EV_PIP_COND = 1,
+};
+
+enum br_ctx_signal_state {
+	BR_CSS_NONE = 2,
+};
+
+#define OP_BBYTE_BASE	0x0c800000000ULL
+#define OP_BB_A_SRC	0x000000000ffULL
+#define OP_BB_BYTE	0x00000000300ULL
+#define OP_BB_B_SRC	0x0000003fc00ULL
+#define OP_BB_I8	0x00000040000ULL
+#define OP_BB_EQ	0x00000080000ULL
+#define OP_BB_DEFBR	0x00000300000ULL
+#define OP_BB_ADDR_LO	0x007ffc00000ULL
+#define OP_BB_ADDR_HI	0x10000000000ULL
+
+#define OP_BALU_BASE	0x0e800000000ULL
+#define OP_BA_A_SRC	0x000000003ffULL
+#define OP_BA_B_SRC	0x000000ffc00ULL
+#define OP_BA_DEFBR	0x00000300000ULL
+#define OP_BA_ADDR_HI	0x0007fc00000ULL
+
+#define OP_IMMED_A_SRC	0x000000003ffULL
+#define OP_IMMED_B_SRC	0x000000ffc00ULL
+#define OP_IMMED_IMM	0x0000ff00000ULL
+#define OP_IMMED_WIDTH	0x00060000000ULL
+#define OP_IMMED_INV	0x00080000000ULL
+#define OP_IMMED_SHIFT	0x00600000000ULL
+#define OP_IMMED_BASE	0x0f000000000ULL
+#define OP_IMMED_WR_AB	0x20000000000ULL
+
+enum immed_width {
+	IMMED_WIDTH_ALL = 0,
+	IMMED_WIDTH_BYTE = 1,
+	IMMED_WIDTH_WORD = 2,
+};
+
+enum immed_shift {
+	IMMED_SHIFT_0B = 0,
+	IMMED_SHIFT_1B = 1,
+	IMMED_SHIFT_2B = 2,
+};
+
+#define OP_SHF_BASE	0x08000000000ULL
+#define OP_SHF_A_SRC	0x000000000ffULL
+#define OP_SHF_SC	0x00000000300ULL
+#define OP_SHF_B_SRC	0x0000003fc00ULL
+#define OP_SHF_I8	0x00000040000ULL
+#define OP_SHF_SW	0x00000080000ULL
+#define OP_SHF_DST	0x0000ff00000ULL
+#define OP_SHF_SHIFT	0x001f0000000ULL
+#define OP_SHF_OP	0x00e00000000ULL
+#define OP_SHF_DST_AB	0x01000000000ULL
+#define OP_SHF_WR_AB	0x20000000000ULL
+
+enum shf_op {
+	SHF_OP_NONE = 0,
+	SHF_OP_AND = 2,
+	SHF_OP_OR = 5,
+};
+
+enum shf_sc {
+	SHF_SC_R_ROT = 0,
+	SHF_SC_R_SHF = 1,
+	SHF_SC_L_SHF = 2,
+	SHF_SC_R_DSHF = 3,
+};
+
+#define OP_ALU_A_SRC	0x000000003ffULL
+#define OP_ALU_B_SRC	0x000000ffc00ULL
+#define OP_ALU_DST	0x0003ff00000ULL
+#define OP_ALU_SW	0x00040000000ULL
+#define OP_ALU_OP	0x00f80000000ULL
+#define OP_ALU_DST_AB	0x01000000000ULL
+#define OP_ALU_BASE	0x0a000000000ULL
+#define OP_ALU_WR_AB	0x20000000000ULL
+
+enum alu_op {
+	ALU_OP_NONE	= 0x00,
+	ALU_OP_ADD	= 0x01,
+	ALU_OP_NEG	= 0x04,
+	ALU_OP_AND	= 0x08,
+	ALU_OP_SUB_C	= 0x0d,
+	ALU_OP_ADD_C	= 0x11,
+	ALU_OP_OR	= 0x14,
+	ALU_OP_SUB	= 0x15,
+	ALU_OP_XOR	= 0x18,
+};
+
+enum alu_dst_ab {
+	ALU_DST_A = 0,
+	ALU_DST_B = 1,
+};
+
+#define OP_LDF_BASE	0x0c000000000ULL
+#define OP_LDF_A_SRC	0x000000000ffULL
+#define OP_LDF_SC	0x00000000300ULL
+#define OP_LDF_B_SRC	0x0000003fc00ULL
+#define OP_LDF_I8	0x00000040000ULL
+#define OP_LDF_SW	0x00000080000ULL
+#define OP_LDF_ZF	0x00000100000ULL
+#define OP_LDF_BMASK	0x0000f000000ULL
+#define OP_LDF_SHF	0x001f0000000ULL
+#define OP_LDF_WR_AB	0x20000000000ULL
+
+#define OP_CMD_A_SRC	 0x000000000ffULL
+#define OP_CMD_CTX	 0x00000000300ULL
+#define OP_CMD_B_SRC	 0x0000003fc00ULL
+#define OP_CMD_TOKEN	 0x000000c0000ULL
+#define OP_CMD_XFER	 0x00001f00000ULL
+#define OP_CMD_CNT	 0x0000e000000ULL
+#define OP_CMD_SIG	 0x000f0000000ULL
+#define OP_CMD_TGT_CMD	 0x07f00000000ULL
+#define OP_CMD_MODE	0x1c0000000000ULL
+
+struct cmd_tgt_act {
+	u8 token;
+	u8 tgt_cmd;
+};
+
+enum cmd_tgt_map {
+	CMD_TGT_READ8,
+	CMD_TGT_WRITE8,
+	CMD_TGT_READ_LE,
+	CMD_TGT_READ_SWAP_LE,
+	__CMD_TGT_MAP_SIZE,
+};
+
+enum cmd_mode {
+	CMD_MODE_40b_AB	= 0,
+	CMD_MODE_40b_BA	= 1,
+	CMD_MODE_32b	= 4,
+};
+
+enum cmd_ctx_swap {
+	CMD_CTX_SWAP = 0,
+	CMD_CTX_NO_SWAP = 3,
+};
+
+#define OP_LCSR_BASE	0x0fc00000000ULL
+#define OP_LCSR_A_SRC	0x000000003ffULL
+#define OP_LCSR_B_SRC	0x000000ffc00ULL
+#define OP_LCSR_WRITE	0x00000200000ULL
+#define OP_LCSR_ADDR	0x001ffc00000ULL
+
+enum lcsr_wr_src {
+	LCSR_WR_AREG,
+	LCSR_WR_BREG,
+	LCSR_WR_IMM,
+};
+
+#define OP_CARB_BASE	0x0e000000000ULL
+#define OP_CARB_OR	0x00000010000ULL
+
+#endif
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_bpf.h b/drivers/net/ethernet/netronome/nfp/nfp_bpf.h
new file mode 100644
index 000000000000..af43e058be97
--- /dev/null
+++ b/drivers/net/ethernet/netronome/nfp/nfp_bpf.h
@@ -0,0 +1,208 @@
+/*
+ * 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.
+ */
+
+#ifndef __NFP_BPF_H__
+#define __NFP_BPF_H__ 1
+
+#include <linux/bpf.h>
+#include <linux/bitfield.h>
+#include <linux/list.h>
+#include <linux/types.h>
+
+#define FIELD_FIT(mask, val)  (!((((u64)val) << __bf_shf(mask)) & ~(mask)))
+
+/* For branch fixup logic use up-most byte of branch instruction as scratch
+ * area.  Remember to clear this before sending instructions to HW!
+ */
+#define OP_BR_SPECIAL	0xff00000000000000ULL
+
+enum br_special {
+	OP_BR_NORMAL = 0,
+	OP_BR_GO_OUT,
+	OP_BR_GO_ABORT,
+};
+
+enum static_regs {
+	STATIC_REG_PKT		= 1,
+#define REG_PKT_BANK	ALU_DST_A
+	STATIC_REG_IMM		= 2, /* Bank AB */
+};
+
+enum nfp_bpf_action_type {
+	NN_ACT_TC_DROP,
+};
+
+/* Software register representation, hardware encoding in asm.h */
+#define NN_REG_TYPE	GENMASK(31, 24)
+#define NN_REG_VAL	GENMASK(7, 0)
+
+enum nfp_bpf_reg_type {
+	NN_REG_GPR_A =	BIT(0),
+	NN_REG_GPR_B =	BIT(1),
+	NN_REG_NNR =	BIT(2),
+	NN_REG_XFER =	BIT(3),
+	NN_REG_IMM =	BIT(4),
+	NN_REG_NONE =	BIT(5),
+};
+
+#define NN_REG_GPR_BOTH	(NN_REG_GPR_A | NN_REG_GPR_B)
+
+#define reg_both(x)	((x) | FIELD_PREP(NN_REG_TYPE, NN_REG_GPR_BOTH))
+#define reg_a(x)	((x) | FIELD_PREP(NN_REG_TYPE, NN_REG_GPR_A))
+#define reg_b(x)	((x) | FIELD_PREP(NN_REG_TYPE, NN_REG_GPR_B))
+#define reg_nnr(x)	((x) | FIELD_PREP(NN_REG_TYPE, NN_REG_NNR))
+#define reg_xfer(x)	((x) | FIELD_PREP(NN_REG_TYPE, NN_REG_XFER))
+#define reg_imm(x)	((x) | FIELD_PREP(NN_REG_TYPE, NN_REG_IMM))
+#define reg_none()	(FIELD_PREP(NN_REG_TYPE, NN_REG_NONE))
+
+#define pkt_reg(np)	reg_a((np)->regs_per_thread - STATIC_REG_PKT)
+#define imm_a(np)	reg_a((np)->regs_per_thread - STATIC_REG_IMM)
+#define imm_b(np)	reg_b((np)->regs_per_thread - STATIC_REG_IMM)
+#define imm_both(np)	reg_both((np)->regs_per_thread - STATIC_REG_IMM)
+
+#define NFP_BPF_ABI_FLAGS	reg_nnr(0)
+#define NFP_BPF_ABI_PKT		reg_nnr(2)
+#define NFP_BPF_ABI_LEN		reg_nnr(3)
+
+struct nfp_prog;
+struct nfp_insn_meta;
+typedef int (*instr_cb_t)(struct nfp_prog *, struct nfp_insn_meta *);
+
+#define nfp_prog_first_meta(nfp_prog)					\
+	list_first_entry(&(nfp_prog)->insns, struct nfp_insn_meta, l)
+#define nfp_prog_last_meta(nfp_prog)					\
+	list_last_entry(&(nfp_prog)->insns, struct nfp_insn_meta, l)
+#define nfp_meta_next(meta)	list_next_entry(meta, l)
+#define nfp_meta_prev(meta)	list_prev_entry(meta, l)
+
+/**
+ * struct nfp_insn_meta - BPF instruction wrapper
+ * @insn: BPF instruction
+ * @off: index of first generated machine instruction (in nfp_prog.prog)
+ * @n: eBPF instruction number
+ * @skip: skip this instruction (optimized out)
+ * @double_cb: callback for second part of the instruction
+ * @l: link on nfp_prog->insns list
+ */
+struct nfp_insn_meta {
+	struct bpf_insn insn;
+	unsigned int off;
+	unsigned short n;
+	bool skip;
+	instr_cb_t double_cb;
+
+	struct list_head l;
+};
+
+#define BPF_SIZE_MASK	0x18
+
+static inline u8 mbpf_class(const struct nfp_insn_meta *meta)
+{
+	return BPF_CLASS(meta->insn.code);
+}
+
+static inline u8 mbpf_src(const struct nfp_insn_meta *meta)
+{
+	return BPF_SRC(meta->insn.code);
+}
+
+static inline u8 mbpf_op(const struct nfp_insn_meta *meta)
+{
+	return BPF_OP(meta->insn.code);
+}
+
+static inline u8 mbpf_mode(const struct nfp_insn_meta *meta)
+{
+	return BPF_MODE(meta->insn.code);
+}
+
+/**
+ * struct nfp_prog - nfp BPF program
+ * @prog: machine code
+ * @prog_len: number of valid instructions in @prog array
+ * @__prog_alloc_len: alloc size of @prog array
+ * @act: BPF program/action type (TC DA, TC with action, XDP etc.)
+ * @num_regs: number of registers used by this program
+ * @regs_per_thread: number of basic registers allocated per thread
+ * @start_off: address of the first instruction in the memory
+ * @tgt_out: jump target for normal exit
+ * @tgt_abort: jump target for abort (e.g. access outside of packet buffer)
+ * @tgt_done: jump target to get the next packet
+ * @n_translated: number of successfully translated instructions (for errors)
+ * @error: error code if something went wrong
+ * @insns: list of BPF instruction wrappers (struct nfp_insn_meta)
+ */
+struct nfp_prog {
+	u64 *prog;
+	unsigned int prog_len;
+	unsigned int __prog_alloc_len;
+
+	enum nfp_bpf_action_type act;
+
+	unsigned int num_regs;
+	unsigned int regs_per_thread;
+
+	unsigned int start_off;
+	unsigned int tgt_out;
+	unsigned int tgt_abort;
+	unsigned int tgt_done;
+
+	unsigned int n_translated;
+	int error;
+
+	struct list_head insns;
+};
+
+struct nfp_bpf_result {
+	unsigned int n_instr;
+	bool dense_mode;
+};
+
+#ifdef CONFIG_BPF_SYSCALL
+int
+nfp_bpf_jit(struct bpf_prog *filter, void *prog, enum nfp_bpf_action_type act,
+	    unsigned int prog_start, unsigned int prog_done,
+	    unsigned int prog_sz, struct nfp_bpf_result *res);
+#else
+int
+nfp_bpf_jit(struct bpf_prog *filter, void *prog, enum nfp_bpf_action_type act,
+	    unsigned int prog_start, unsigned int prog_done,
+	    unsigned int prog_sz, struct nfp_bpf_result *res)
+{
+	return -ENOTSUPP;
+}
+#endif
+
+int nfp_prog_verify(struct nfp_prog *nfp_prog, struct bpf_prog *prog);
+
+#endif
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c b/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
new file mode 100644
index 000000000000..dfcf162ccbb8
--- /dev/null
+++ b/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
@@ -0,0 +1,1729 @@
+/*
+ * 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.
+ */
+
+#define pr_fmt(fmt)	"NFP net bpf: " fmt
+
+#include <linux/kernel.h>
+#include <linux/bpf.h>
+#include <linux/filter.h>
+#include <linux/pkt_cls.h>
+#include <linux/unistd.h>
+
+#include "nfp_asm.h"
+#include "nfp_bpf.h"
+
+/* --- NFP prog --- */
+/* Foreach "multiple" entries macros provide pos and next<n> pointers.
+ * It's safe to modify the next pointers (but not pos).
+ */
+#define nfp_for_each_insn_walk2(nfp_prog, pos, next)			\
+	for (pos = list_first_entry(&(nfp_prog)->insns, typeof(*pos), l), \
+	     next = list_next_entry(pos, l);			\
+	     &(nfp_prog)->insns != &pos->l &&			\
+	     &(nfp_prog)->insns != &next->l;			\
+	     pos = nfp_meta_next(pos),				\
+	     next = nfp_meta_next(pos))
+
+#define nfp_for_each_insn_walk3(nfp_prog, pos, next, next2)		\
+	for (pos = list_first_entry(&(nfp_prog)->insns, typeof(*pos), l), \
+	     next = list_next_entry(pos, l),			\
+	     next2 = list_next_entry(next, l);			\
+	     &(nfp_prog)->insns != &pos->l &&			\
+	     &(nfp_prog)->insns != &next->l &&			\
+	     &(nfp_prog)->insns != &next2->l;			\
+	     pos = nfp_meta_next(pos),				\
+	     next = nfp_meta_next(pos),				\
+	     next2 = nfp_meta_next(next))
+
+static bool
+nfp_meta_has_next(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return meta->l.next != &nfp_prog->insns;
+}
+
+static bool
+nfp_meta_has_prev(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return meta->l.prev != &nfp_prog->insns;
+}
+
+static void nfp_prog_free(struct nfp_prog *nfp_prog)
+{
+	struct nfp_insn_meta *meta, *tmp;
+
+	list_for_each_entry_safe(meta, tmp, &nfp_prog->insns, l) {
+		list_del(&meta->l);
+		kfree(meta);
+	}
+	kfree(nfp_prog);
+}
+
+static void nfp_prog_push(struct nfp_prog *nfp_prog, u64 insn)
+{
+	if (nfp_prog->__prog_alloc_len == nfp_prog->prog_len) {
+		nfp_prog->error = -ENOSPC;
+		return;
+	}
+
+	nfp_prog->prog[nfp_prog->prog_len] = insn;
+	nfp_prog->prog_len++;
+}
+
+static unsigned int nfp_prog_current_offset(struct nfp_prog *nfp_prog)
+{
+	return nfp_prog->start_off + nfp_prog->prog_len;
+}
+
+static unsigned int
+nfp_prog_offset_to_index(struct nfp_prog *nfp_prog, unsigned int offset)
+{
+	return offset - nfp_prog->start_off;
+}
+
+/* --- SW reg --- */
+struct nfp_insn_ur_regs {
+	enum alu_dst_ab dst_ab;
+	u16 dst;
+	u16 areg, breg;
+	bool swap;
+	bool wr_both;
+};
+
+struct nfp_insn_re_regs {
+	enum alu_dst_ab dst_ab;
+	u8 dst;
+	u8 areg, breg;
+	bool swap;
+	bool wr_both;
+	bool i8;
+};
+
+static u16 nfp_swreg_to_unreg(u32 swreg, bool is_dst)
+{
+	u16 val = FIELD_GET(NN_REG_VAL, swreg);
+
+	switch (FIELD_GET(NN_REG_TYPE, swreg)) {
+	case NN_REG_GPR_A:
+	case NN_REG_GPR_B:
+	case NN_REG_GPR_BOTH:
+		return val;
+	case NN_REG_NNR:
+		return UR_REG_NN | val;
+	case NN_REG_XFER:
+		return UR_REG_XFR | val;
+	case NN_REG_IMM:
+		if (val & ~0xff) {
+			pr_err("immediate too large\n");
+			return 0;
+		}
+		return UR_REG_IMM_encode(val);
+	case NN_REG_NONE:
+		return is_dst ? UR_REG_NO_DST : REG_NONE;
+	default:
+		pr_err("unrecognized reg encoding %08x\n", swreg);
+		return 0;
+	}
+}
+
+static int
+swreg_to_unrestricted(u32 dst, u32 lreg, u32 rreg, struct nfp_insn_ur_regs *reg)
+{
+	memset(reg, 0, sizeof(*reg));
+
+	/* Decode destination */
+	if (FIELD_GET(NN_REG_TYPE, dst) == NN_REG_IMM)
+		return -EFAULT;
+
+	if (FIELD_GET(NN_REG_TYPE, dst) == NN_REG_GPR_B)
+		reg->dst_ab = ALU_DST_B;
+	if (FIELD_GET(NN_REG_TYPE, dst) == NN_REG_GPR_BOTH)
+		reg->wr_both = true;
+	reg->dst = nfp_swreg_to_unreg(dst, true);
+
+	/* Decode source operands */
+	if (FIELD_GET(NN_REG_TYPE, lreg) == FIELD_GET(NN_REG_TYPE, rreg))
+		return -EFAULT;
+
+	if (FIELD_GET(NN_REG_TYPE, lreg) == NN_REG_GPR_B ||
+	    FIELD_GET(NN_REG_TYPE, rreg) == NN_REG_GPR_A) {
+		reg->areg = nfp_swreg_to_unreg(rreg, false);
+		reg->breg = nfp_swreg_to_unreg(lreg, false);
+		reg->swap = true;
+	} else {
+		reg->areg = nfp_swreg_to_unreg(lreg, false);
+		reg->breg = nfp_swreg_to_unreg(rreg, false);
+	}
+
+	return 0;
+}
+
+static u16 nfp_swreg_to_rereg(u32 swreg, bool is_dst, bool has_imm8, bool *i8)
+{
+	u16 val = FIELD_GET(NN_REG_VAL, swreg);
+
+	switch (FIELD_GET(NN_REG_TYPE, swreg)) {
+	case NN_REG_GPR_A:
+	case NN_REG_GPR_B:
+	case NN_REG_GPR_BOTH:
+		return val;
+	case NN_REG_XFER:
+		return RE_REG_XFR | val;
+	case NN_REG_IMM:
+		if (val & ~(0x7f | has_imm8 << 7)) {
+			pr_err("immediate too large\n");
+			return 0;
+		}
+		*i8 = val & 0x80;
+		return RE_REG_IMM_encode(val & 0x7f);
+	case NN_REG_NONE:
+		return is_dst ? RE_REG_NO_DST : REG_NONE;
+	default:
+		pr_err("unrecognized reg encoding\n");
+		return 0;
+	}
+}
+
+static int
+swreg_to_restricted(u32 dst, u32 lreg, u32 rreg, struct nfp_insn_re_regs *reg,
+		    bool has_imm8)
+{
+	memset(reg, 0, sizeof(*reg));
+
+	/* Decode destination */
+	if (FIELD_GET(NN_REG_TYPE, dst) == NN_REG_IMM)
+		return -EFAULT;
+
+	if (FIELD_GET(NN_REG_TYPE, dst) == NN_REG_GPR_B)
+		reg->dst_ab = ALU_DST_B;
+	if (FIELD_GET(NN_REG_TYPE, dst) == NN_REG_GPR_BOTH)
+		reg->wr_both = true;
+	reg->dst = nfp_swreg_to_rereg(dst, true, false, NULL);
+
+	/* Decode source operands */
+	if (FIELD_GET(NN_REG_TYPE, lreg) == FIELD_GET(NN_REG_TYPE, rreg))
+		return -EFAULT;
+
+	if (FIELD_GET(NN_REG_TYPE, lreg) == NN_REG_GPR_B ||
+	    FIELD_GET(NN_REG_TYPE, rreg) == NN_REG_GPR_A) {
+		reg->areg = nfp_swreg_to_rereg(rreg, false, has_imm8, &reg->i8);
+		reg->breg = nfp_swreg_to_rereg(lreg, false, has_imm8, &reg->i8);
+		reg->swap = true;
+	} else {
+		reg->areg = nfp_swreg_to_rereg(lreg, false, has_imm8, &reg->i8);
+		reg->breg = nfp_swreg_to_rereg(rreg, false, has_imm8, &reg->i8);
+	}
+
+	return 0;
+}
+
+/* --- Emitters --- */
+static const struct cmd_tgt_act cmd_tgt_act[__CMD_TGT_MAP_SIZE] = {
+	[CMD_TGT_WRITE8] =		{ 0x00, 0x42 },
+	[CMD_TGT_READ8] =		{ 0x01, 0x43 },
+	[CMD_TGT_READ_LE] =		{ 0x01, 0x40 },
+	[CMD_TGT_READ_SWAP_LE] =	{ 0x03, 0x40 },
+};
+
+static void
+__emit_cmd(struct nfp_prog *nfp_prog, enum cmd_tgt_map op,
+	   u8 mode, u8 xfer, u8 areg, u8 breg, u8 size, bool sync)
+{
+	enum cmd_ctx_swap ctx;
+	u64 insn;
+
+	if (sync)
+		ctx = CMD_CTX_SWAP;
+	else
+		ctx = CMD_CTX_NO_SWAP;
+
+	insn =	FIELD_PREP(OP_CMD_A_SRC, areg) |
+		FIELD_PREP(OP_CMD_CTX, ctx) |
+		FIELD_PREP(OP_CMD_B_SRC, breg) |
+		FIELD_PREP(OP_CMD_TOKEN, cmd_tgt_act[op].token) |
+		FIELD_PREP(OP_CMD_XFER, xfer) |
+		FIELD_PREP(OP_CMD_CNT, size) |
+		FIELD_PREP(OP_CMD_SIG, sync) |
+		FIELD_PREP(OP_CMD_TGT_CMD, cmd_tgt_act[op].tgt_cmd) |
+		FIELD_PREP(OP_CMD_MODE, mode);
+
+	nfp_prog_push(nfp_prog, insn);
+}
+
+static void
+emit_cmd(struct nfp_prog *nfp_prog, enum cmd_tgt_map op,
+	 u8 mode, u8 xfer, u32 lreg, u32 rreg, u8 size, bool sync)
+{
+	struct nfp_insn_re_regs reg;
+	int err;
+
+	err = swreg_to_restricted(reg_none(), lreg, rreg, &reg, false);
+	if (err) {
+		nfp_prog->error = err;
+		return;
+	}
+	if (reg.swap) {
+		pr_err("cmd can't swap arguments\n");
+		nfp_prog->error = -EFAULT;
+		return;
+	}
+
+	__emit_cmd(nfp_prog, op, mode, xfer, reg.areg, reg.breg, size, sync);
+}
+
+static void
+__emit_br(struct nfp_prog *nfp_prog, enum br_mask mask, enum br_ev_pip ev_pip,
+	  enum br_ctx_signal_state css, u16 addr, u8 defer)
+{
+	u16 addr_lo, addr_hi;
+	u64 insn;
+
+	addr_lo = addr & (OP_BR_ADDR_LO >> __bf_shf(OP_BR_ADDR_LO));
+	addr_hi = addr != addr_lo;
+
+	insn = OP_BR_BASE |
+		FIELD_PREP(OP_BR_MASK, mask) |
+		FIELD_PREP(OP_BR_EV_PIP, ev_pip) |
+		FIELD_PREP(OP_BR_CSS, css) |
+		FIELD_PREP(OP_BR_DEFBR, defer) |
+		FIELD_PREP(OP_BR_ADDR_LO, addr_lo) |
+		FIELD_PREP(OP_BR_ADDR_HI, addr_hi);
+
+	nfp_prog_push(nfp_prog, insn);
+}
+
+static void
+emit_br(struct nfp_prog *nfp_prog, enum br_mask mask, u16 addr, u8 defer)
+{
+	__emit_br(nfp_prog, mask,
+		  mask != BR_UNC ? BR_EV_PIP_COND : BR_EV_PIP_UNCOND,
+		  BR_CSS_NONE, addr, defer);
+}
+
+static void
+__emit_br_byte(struct nfp_prog *nfp_prog, u8 areg, u8 breg, bool imm8,
+	       u8 byte, bool equal, u16 addr, u8 defer)
+{
+	u16 addr_lo, addr_hi;
+	u64 insn;
+
+	addr_lo = addr & (OP_BB_ADDR_LO >> __bf_shf(OP_BB_ADDR_LO));
+	addr_hi = addr != addr_lo;
+
+	insn = OP_BBYTE_BASE |
+		FIELD_PREP(OP_BB_A_SRC, areg) |
+		FIELD_PREP(OP_BB_BYTE, byte) |
+		FIELD_PREP(OP_BB_B_SRC, breg) |
+		FIELD_PREP(OP_BB_I8, imm8) |
+		FIELD_PREP(OP_BB_EQ, equal) |
+		FIELD_PREP(OP_BB_DEFBR, defer) |
+		FIELD_PREP(OP_BB_ADDR_LO, addr_lo) |
+		FIELD_PREP(OP_BB_ADDR_HI, addr_hi);
+
+	nfp_prog_push(nfp_prog, insn);
+}
+
+static void
+emit_br_byte_neq(struct nfp_prog *nfp_prog,
+		 u32 dst, u8 imm, u8 byte, u16 addr, u8 defer)
+{
+	struct nfp_insn_re_regs reg;
+	int err;
+
+	err = swreg_to_restricted(reg_none(), dst, reg_imm(imm), &reg, true);
+	if (err) {
+		nfp_prog->error = err;
+		return;
+	}
+
+	__emit_br_byte(nfp_prog, reg.areg, reg.breg, reg.i8, byte, false, addr,
+		       defer);
+}
+
+static void
+__emit_immed(struct nfp_prog *nfp_prog, u16 areg, u16 breg, u16 imm_hi,
+	     enum immed_width width, bool invert,
+	     enum immed_shift shift, bool wr_both)
+{
+	u64 insn;
+
+	insn = OP_IMMED_BASE |
+		FIELD_PREP(OP_IMMED_A_SRC, areg) |
+		FIELD_PREP(OP_IMMED_B_SRC, breg) |
+		FIELD_PREP(OP_IMMED_IMM, imm_hi) |
+		FIELD_PREP(OP_IMMED_WIDTH, width) |
+		FIELD_PREP(OP_IMMED_INV, invert) |
+		FIELD_PREP(OP_IMMED_SHIFT, shift) |
+		FIELD_PREP(OP_IMMED_WR_AB, wr_both);
+
+	nfp_prog_push(nfp_prog, insn);
+}
+
+static void
+emit_immed(struct nfp_prog *nfp_prog, u32 dst, u16 imm,
+	   enum immed_width width, bool invert, enum immed_shift shift)
+{
+	struct nfp_insn_ur_regs reg;
+	int err;
+
+	if (FIELD_GET(NN_REG_TYPE, dst) == NN_REG_IMM) {
+		nfp_prog->error = -EFAULT;
+		return;
+	}
+
+	err = swreg_to_unrestricted(dst, dst, reg_imm(imm & 0xff), &reg);
+	if (err) {
+		nfp_prog->error = err;
+		return;
+	}
+
+	__emit_immed(nfp_prog, reg.areg, reg.breg, imm >> 8, width,
+		     invert, shift, reg.wr_both);
+}
+
+static void
+__emit_shf(struct nfp_prog *nfp_prog, u16 dst, enum alu_dst_ab dst_ab,
+	   enum shf_sc sc, u8 shift,
+	   u16 areg, enum shf_op op, u16 breg, bool i8, bool sw, bool wr_both)
+{
+	u64 insn;
+
+	if (!FIELD_FIT(OP_SHF_SHIFT, shift)) {
+		nfp_prog->error = -EFAULT;
+		return;
+	}
+
+	if (sc == SHF_SC_L_SHF)
+		shift = 32 - shift;
+
+	insn = OP_SHF_BASE |
+		FIELD_PREP(OP_SHF_A_SRC, areg) |
+		FIELD_PREP(OP_SHF_SC, sc) |
+		FIELD_PREP(OP_SHF_B_SRC, breg) |
+		FIELD_PREP(OP_SHF_I8, i8) |
+		FIELD_PREP(OP_SHF_SW, sw) |
+		FIELD_PREP(OP_SHF_DST, dst) |
+		FIELD_PREP(OP_SHF_SHIFT, shift) |
+		FIELD_PREP(OP_SHF_OP, op) |
+		FIELD_PREP(OP_SHF_DST_AB, dst_ab) |
+		FIELD_PREP(OP_SHF_WR_AB, wr_both);
+
+	nfp_prog_push(nfp_prog, insn);
+}
+
+static void
+emit_shf(struct nfp_prog *nfp_prog, u32 dst, u32 lreg, enum shf_op op, u32 rreg,
+	 enum shf_sc sc, u8 shift)
+{
+	struct nfp_insn_re_regs reg;
+	int err;
+
+	err = swreg_to_restricted(dst, lreg, rreg, &reg, true);
+	if (err) {
+		nfp_prog->error = err;
+		return;
+	}
+
+	__emit_shf(nfp_prog, reg.dst, reg.dst_ab, sc, shift,
+		   reg.areg, op, reg.breg, reg.i8, reg.swap, reg.wr_both);
+}
+
+static void
+__emit_alu(struct nfp_prog *nfp_prog, u16 dst, enum alu_dst_ab dst_ab,
+	   u16 areg, enum alu_op op, u16 breg, bool swap, bool wr_both)
+{
+	u64 insn;
+
+	insn = OP_ALU_BASE |
+		FIELD_PREP(OP_ALU_A_SRC, areg) |
+		FIELD_PREP(OP_ALU_B_SRC, breg) |
+		FIELD_PREP(OP_ALU_DST, dst) |
+		FIELD_PREP(OP_ALU_SW, swap) |
+		FIELD_PREP(OP_ALU_OP, op) |
+		FIELD_PREP(OP_ALU_DST_AB, dst_ab) |
+		FIELD_PREP(OP_ALU_WR_AB, wr_both);
+
+	nfp_prog_push(nfp_prog, insn);
+}
+
+static void
+emit_alu(struct nfp_prog *nfp_prog, u32 dst, u32 lreg, enum alu_op op, u32 rreg)
+{
+	struct nfp_insn_ur_regs reg;
+	int err;
+
+	err = swreg_to_unrestricted(dst, lreg, rreg, &reg);
+	if (err) {
+		nfp_prog->error = err;
+		return;
+	}
+
+	__emit_alu(nfp_prog, reg.dst, reg.dst_ab,
+		   reg.areg, op, reg.breg, reg.swap, reg.wr_both);
+}
+
+static void
+__emit_ld_field(struct nfp_prog *nfp_prog, enum shf_sc sc,
+		u8 areg, u8 bmask, u8 breg, u8 shift, bool imm8,
+		bool zero, bool swap, bool wr_both)
+{
+	u64 insn;
+
+	insn = OP_LDF_BASE |
+		FIELD_PREP(OP_LDF_A_SRC, areg) |
+		FIELD_PREP(OP_LDF_SC, sc) |
+		FIELD_PREP(OP_LDF_B_SRC, breg) |
+		FIELD_PREP(OP_LDF_I8, imm8) |
+		FIELD_PREP(OP_LDF_SW, swap) |
+		FIELD_PREP(OP_LDF_ZF, zero) |
+		FIELD_PREP(OP_LDF_BMASK, bmask) |
+		FIELD_PREP(OP_LDF_SHF, shift) |
+		FIELD_PREP(OP_LDF_WR_AB, wr_both);
+
+	nfp_prog_push(nfp_prog, insn);
+}
+
+static void
+emit_ld_field_any(struct nfp_prog *nfp_prog, enum shf_sc sc, u8 shift,
+		  u32 dst, u8 bmask, u32 src, bool zero)
+{
+	struct nfp_insn_re_regs reg;
+	int err;
+
+	err = swreg_to_restricted(reg_none(), dst, src, &reg, true);
+	if (err) {
+		nfp_prog->error = err;
+		return;
+	}
+
+	__emit_ld_field(nfp_prog, sc, reg.areg, bmask, reg.breg, shift,
+			reg.i8, zero, reg.swap, reg.wr_both);
+}
+
+static void
+emit_ld_field(struct nfp_prog *nfp_prog, u32 dst, u8 bmask, u32 src,
+	      enum shf_sc sc, u8 shift)
+{
+	emit_ld_field_any(nfp_prog, sc, shift, dst, bmask, src, false);
+}
+
+/* --- Wrappers --- */
+static bool pack_immed(u32 imm, u16 *val, enum immed_shift *shift)
+{
+	if (!(imm & 0xffff0000)) {
+		*val = imm;
+		*shift = IMMED_SHIFT_0B;
+	} else if (!(imm & 0xff0000ff)) {
+		*val = imm >> 8;
+		*shift = IMMED_SHIFT_1B;
+	} else if (!(imm & 0x0000ffff)) {
+		*val = imm >> 16;
+		*shift = IMMED_SHIFT_2B;
+	} else {
+		return false;
+	}
+
+	return true;
+}
+
+static void wrp_immed(struct nfp_prog *nfp_prog, u32 dst, u32 imm)
+{
+	enum immed_shift shift;
+	u16 val;
+
+	if (pack_immed(imm, &val, &shift)) {
+		emit_immed(nfp_prog, dst, val, IMMED_WIDTH_ALL, false, shift);
+	} else if (pack_immed(~imm, &val, &shift)) {
+		emit_immed(nfp_prog, dst, val, IMMED_WIDTH_ALL, true, shift);
+	} else {
+		emit_immed(nfp_prog, dst, imm & 0xffff, IMMED_WIDTH_ALL,
+			   false, IMMED_SHIFT_0B);
+		emit_immed(nfp_prog, dst, imm >> 16, IMMED_WIDTH_WORD,
+			   false, IMMED_SHIFT_2B);
+	}
+}
+
+/* ur_load_imm_any() - encode immediate or use tmp register (unrestricted)
+ * If the @imm is small enough encode it directly in operand and return
+ * otherwise load @imm to a spare register and return its encoding.
+ */
+static u32 ur_load_imm_any(struct nfp_prog *nfp_prog, u32 imm, u32 tmp_reg)
+{
+	if (FIELD_FIT(UR_REG_IMM_MAX, imm))
+		return reg_imm(imm);
+
+	wrp_immed(nfp_prog, tmp_reg, imm);
+	return tmp_reg;
+}
+
+/* re_load_imm_any() - encode immediate or use tmp register (restricted)
+ * If the @imm is small enough encode it directly in operand and return
+ * otherwise load @imm to a spare register and return its encoding.
+ */
+static u32 re_load_imm_any(struct nfp_prog *nfp_prog, u32 imm, u32 tmp_reg)
+{
+	if (FIELD_FIT(RE_REG_IMM_MAX, imm))
+		return reg_imm(imm);
+
+	wrp_immed(nfp_prog, tmp_reg, imm);
+	return tmp_reg;
+}
+
+static void
+wrp_br_special(struct nfp_prog *nfp_prog, enum br_mask mask,
+	       enum br_special special)
+{
+	emit_br(nfp_prog, mask, 0, 0);
+
+	nfp_prog->prog[nfp_prog->prog_len - 1] |=
+		FIELD_PREP(OP_BR_SPECIAL, special);
+}
+
+static void wrp_reg_mov(struct nfp_prog *nfp_prog, u16 dst, u16 src)
+{
+	emit_alu(nfp_prog, reg_both(dst), reg_none(), ALU_OP_NONE, reg_b(src));
+}
+
+static int
+construct_data_ind_ld(struct nfp_prog *nfp_prog, u16 offset,
+		      u16 src, bool src_valid, u8 size)
+{
+	unsigned int i;
+	u16 shift, sz;
+	u32 tmp_reg;
+
+	/* We load the value from the address indicated in @offset and then
+	 * shift out the data we don't need.  Note: this is big endian!
+	 */
+	sz = size < 4 ? 4 : size;
+	shift = size < 4 ? 4 - size : 0;
+
+	if (src_valid) {
+		/* Calculate the true offset (src_reg + imm) */
+		tmp_reg = ur_load_imm_any(nfp_prog, offset, imm_b(nfp_prog));
+		emit_alu(nfp_prog, imm_both(nfp_prog),
+			 reg_a(src), ALU_OP_ADD, tmp_reg);
+		/* Check packet length (size guaranteed to fit b/c it's u8) */
+		emit_alu(nfp_prog, imm_a(nfp_prog),
+			 imm_a(nfp_prog), ALU_OP_ADD, reg_imm(size));
+		emit_alu(nfp_prog, reg_none(),
+			 NFP_BPF_ABI_LEN, ALU_OP_SUB, imm_a(nfp_prog));
+		wrp_br_special(nfp_prog, BR_BLO, OP_BR_GO_ABORT);
+		/* Load data */
+		emit_cmd(nfp_prog, CMD_TGT_READ8, CMD_MODE_32b, 0,
+			 pkt_reg(nfp_prog), imm_b(nfp_prog), sz - 1, true);
+	} else {
+		/* Check packet length */
+		tmp_reg = ur_load_imm_any(nfp_prog, offset + size,
+					  imm_a(nfp_prog));
+		emit_alu(nfp_prog, reg_none(),
+			 NFP_BPF_ABI_LEN, ALU_OP_SUB, tmp_reg);
+		wrp_br_special(nfp_prog, BR_BLO, OP_BR_GO_ABORT);
+		/* Load data */
+		tmp_reg = re_load_imm_any(nfp_prog, offset, imm_b(nfp_prog));
+		emit_cmd(nfp_prog, CMD_TGT_READ8, CMD_MODE_32b, 0,
+			 pkt_reg(nfp_prog), tmp_reg, sz - 1, true);
+	}
+
+	i = 0;
+	if (shift)
+		emit_shf(nfp_prog, reg_both(0), reg_none(), SHF_OP_NONE,
+			 reg_xfer(0), SHF_SC_R_SHF, shift * 8);
+	else
+		for (; i * 4 < size; i++)
+			emit_alu(nfp_prog, reg_both(i),
+				 reg_none(), ALU_OP_NONE, reg_xfer(i));
+
+	if (i < 2)
+		wrp_immed(nfp_prog, reg_both(1), 0);
+
+	return 0;
+}
+
+static int construct_data_ld(struct nfp_prog *nfp_prog, u16 offset, u8 size)
+{
+	return construct_data_ind_ld(nfp_prog, offset, 0, false, size);
+}
+
+static void
+wrp_alu_imm(struct nfp_prog *nfp_prog, u8 dst, enum alu_op alu_op, u32 imm)
+{
+	u32 tmp_reg;
+
+	if (alu_op == ALU_OP_AND) {
+		if (!imm)
+			wrp_immed(nfp_prog, reg_both(dst), 0);
+		if (!imm || !~imm)
+			return;
+	}
+	if (alu_op == ALU_OP_OR) {
+		if (!~imm)
+			wrp_immed(nfp_prog, reg_both(dst), ~0U);
+		if (!imm || !~imm)
+			return;
+	}
+	if (alu_op == ALU_OP_XOR) {
+		if (!~imm)
+			emit_alu(nfp_prog, reg_both(dst), reg_none(),
+				 ALU_OP_NEG, reg_b(dst));
+		if (!imm || !~imm)
+			return;
+	}
+
+	tmp_reg = ur_load_imm_any(nfp_prog, imm, imm_b(nfp_prog));
+	emit_alu(nfp_prog, reg_both(dst), reg_a(dst), alu_op, tmp_reg);
+}
+
+static int
+wrp_alu64_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
+	      enum alu_op alu_op, bool skip)
+{
+	const struct bpf_insn *insn = &meta->insn;
+	u64 imm = insn->imm; /* sign extend */
+
+	if (skip) {
+		meta->skip = true;
+		return 0;
+	}
+
+	wrp_alu_imm(nfp_prog, insn->dst_reg * 2, alu_op, imm & ~0U);
+	wrp_alu_imm(nfp_prog, insn->dst_reg * 2 + 1, alu_op, imm >> 32);
+
+	return 0;
+}
+
+static int
+wrp_alu64_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
+	      enum alu_op alu_op)
+{
+	u8 dst = meta->insn.dst_reg * 2, src = meta->insn.src_reg * 2;
+
+	emit_alu(nfp_prog, reg_both(dst), reg_a(dst), alu_op, reg_b(src));
+	emit_alu(nfp_prog, reg_both(dst + 1),
+		 reg_a(dst + 1), alu_op, reg_b(src + 1));
+
+	return 0;
+}
+
+static int
+wrp_alu32_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
+	      enum alu_op alu_op, bool skip)
+{
+	const struct bpf_insn *insn = &meta->insn;
+
+	if (skip) {
+		meta->skip = true;
+		return 0;
+	}
+
+	wrp_alu_imm(nfp_prog, insn->dst_reg * 2, alu_op, insn->imm);
+	wrp_immed(nfp_prog, reg_both(insn->dst_reg * 2 + 1), 0);
+
+	return 0;
+}
+
+static int
+wrp_alu32_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
+	      enum alu_op alu_op)
+{
+	u8 dst = meta->insn.dst_reg * 2, src = meta->insn.src_reg * 2;
+
+	emit_alu(nfp_prog, reg_both(dst), reg_a(dst), alu_op, reg_b(src));
+	wrp_immed(nfp_prog, reg_both(meta->insn.dst_reg * 2 + 1), 0);
+
+	return 0;
+}
+
+static void
+wrp_test_reg_one(struct nfp_prog *nfp_prog, u8 dst, enum alu_op alu_op, u8 src,
+		 enum br_mask br_mask, u16 off)
+{
+	emit_alu(nfp_prog, reg_none(), reg_a(dst), alu_op, reg_b(src));
+	emit_br(nfp_prog, br_mask, off, 0);
+}
+
+static int
+wrp_test_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
+	     enum alu_op alu_op, enum br_mask br_mask)
+{
+	const struct bpf_insn *insn = &meta->insn;
+
+	if (insn->off < 0) /* TODO */
+		return -ENOTSUPP;
+
+	wrp_test_reg_one(nfp_prog, insn->dst_reg * 2, alu_op,
+			 insn->src_reg * 2, br_mask, insn->off);
+	wrp_test_reg_one(nfp_prog, insn->dst_reg * 2 + 1, alu_op,
+			 insn->src_reg * 2 + 1, br_mask, insn->off);
+
+	return 0;
+}
+
+static int
+wrp_cmp_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
+	    enum br_mask br_mask, bool swap)
+{
+	const struct bpf_insn *insn = &meta->insn;
+	u64 imm = insn->imm; /* sign extend */
+	u8 reg = insn->dst_reg * 2;
+	u32 tmp_reg;
+
+	if (insn->off < 0) /* TODO */
+		return -ENOTSUPP;
+
+	tmp_reg = ur_load_imm_any(nfp_prog, imm & ~0U, imm_b(nfp_prog));
+	if (!swap)
+		emit_alu(nfp_prog, reg_none(), reg_a(reg), ALU_OP_SUB, tmp_reg);
+	else
+		emit_alu(nfp_prog, reg_none(), tmp_reg, ALU_OP_SUB, reg_a(reg));
+
+	tmp_reg = ur_load_imm_any(nfp_prog, imm >> 32, imm_b(nfp_prog));
+	if (!swap)
+		emit_alu(nfp_prog, reg_none(),
+			 reg_a(reg + 1), ALU_OP_SUB_C, tmp_reg);
+	else
+		emit_alu(nfp_prog, reg_none(),
+			 tmp_reg, ALU_OP_SUB_C, reg_a(reg + 1));
+
+	emit_br(nfp_prog, br_mask, insn->off, 0);
+
+	return 0;
+}
+
+static int
+wrp_cmp_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
+	    enum br_mask br_mask, bool swap)
+{
+	const struct bpf_insn *insn = &meta->insn;
+	u8 areg = insn->src_reg * 2, breg = insn->dst_reg * 2;
+
+	if (insn->off < 0) /* TODO */
+		return -ENOTSUPP;
+
+	if (swap) {
+		areg ^= breg;
+		breg ^= areg;
+		areg ^= breg;
+	}
+
+	emit_alu(nfp_prog, reg_none(), reg_a(areg), ALU_OP_SUB, reg_b(breg));
+	emit_alu(nfp_prog, reg_none(),
+		 reg_a(areg + 1), ALU_OP_SUB_C, reg_b(breg + 1));
+	emit_br(nfp_prog, br_mask, insn->off, 0);
+
+	return 0;
+}
+
+/* --- Callbacks --- */
+static int mov_reg64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+
+	wrp_reg_mov(nfp_prog, insn->dst_reg * 2, insn->src_reg * 2);
+	wrp_reg_mov(nfp_prog, insn->dst_reg * 2 + 1, insn->src_reg * 2 + 1);
+
+	return 0;
+}
+
+static int mov_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	u64 imm = meta->insn.imm; /* sign extend */
+
+	wrp_immed(nfp_prog, reg_both(meta->insn.dst_reg * 2), imm & ~0U);
+	wrp_immed(nfp_prog, reg_both(meta->insn.dst_reg * 2 + 1), imm >> 32);
+
+	return 0;
+}
+
+static int xor_reg64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu64_reg(nfp_prog, meta, ALU_OP_XOR);
+}
+
+static int xor_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu64_imm(nfp_prog, meta, ALU_OP_XOR, !meta->insn.imm);
+}
+
+static int and_reg64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu64_reg(nfp_prog, meta, ALU_OP_AND);
+}
+
+static int and_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu64_imm(nfp_prog, meta, ALU_OP_AND, !~meta->insn.imm);
+}
+
+static int or_reg64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu64_reg(nfp_prog, meta, ALU_OP_OR);
+}
+
+static int or_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu64_imm(nfp_prog, meta, ALU_OP_OR, !meta->insn.imm);
+}
+
+static int add_reg64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+
+	emit_alu(nfp_prog, reg_both(insn->dst_reg * 2),
+		 reg_a(insn->dst_reg * 2), ALU_OP_ADD,
+		 reg_b(insn->src_reg * 2));
+	emit_alu(nfp_prog, reg_both(insn->dst_reg * 2 + 1),
+		 reg_a(insn->dst_reg * 2 + 1), ALU_OP_ADD_C,
+		 reg_b(insn->src_reg * 2 + 1));
+
+	return 0;
+}
+
+static int add_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+	u64 imm = insn->imm; /* sign extend */
+
+	wrp_alu_imm(nfp_prog, insn->dst_reg * 2, ALU_OP_ADD, imm & ~0U);
+	wrp_alu_imm(nfp_prog, insn->dst_reg * 2 + 1, ALU_OP_ADD_C, imm >> 32);
+
+	return 0;
+}
+
+static int sub_reg64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+
+	emit_alu(nfp_prog, reg_both(insn->dst_reg * 2),
+		 reg_a(insn->dst_reg * 2), ALU_OP_SUB,
+		 reg_b(insn->src_reg * 2));
+	emit_alu(nfp_prog, reg_both(insn->dst_reg * 2 + 1),
+		 reg_a(insn->dst_reg * 2 + 1), ALU_OP_SUB_C,
+		 reg_b(insn->src_reg * 2 + 1));
+
+	return 0;
+}
+
+static int sub_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+	u64 imm = insn->imm; /* sign extend */
+
+	wrp_alu_imm(nfp_prog, insn->dst_reg * 2, ALU_OP_SUB, imm & ~0U);
+	wrp_alu_imm(nfp_prog, insn->dst_reg * 2 + 1, ALU_OP_SUB_C, imm >> 32);
+
+	return 0;
+}
+
+static int shl_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+
+	if (insn->imm != 32)
+		return 1; /* TODO */
+
+	wrp_reg_mov(nfp_prog, insn->dst_reg * 2 + 1, insn->dst_reg * 2);
+	wrp_immed(nfp_prog, reg_both(insn->dst_reg * 2), 0);
+
+	return 0;
+}
+
+static int shr_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+
+	if (insn->imm != 32)
+		return 1; /* TODO */
+
+	wrp_reg_mov(nfp_prog, insn->dst_reg * 2, insn->dst_reg * 2 + 1);
+	wrp_immed(nfp_prog, reg_both(insn->dst_reg * 2 + 1), 0);
+
+	return 0;
+}
+
+static int mov_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+
+	wrp_reg_mov(nfp_prog, insn->dst_reg * 2,  insn->src_reg * 2);
+	wrp_immed(nfp_prog, reg_both(insn->dst_reg * 2 + 1), 0);
+
+	return 0;
+}
+
+static int mov_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+
+	wrp_immed(nfp_prog, reg_both(insn->dst_reg * 2), insn->imm);
+	wrp_immed(nfp_prog, reg_both(insn->dst_reg * 2 + 1), 0);
+
+	return 0;
+}
+
+static int xor_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu32_reg(nfp_prog, meta, ALU_OP_XOR);
+}
+
+static int xor_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu32_imm(nfp_prog, meta, ALU_OP_XOR, !~meta->insn.imm);
+}
+
+static int and_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu32_reg(nfp_prog, meta, ALU_OP_AND);
+}
+
+static int and_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu32_imm(nfp_prog, meta, ALU_OP_AND, !~meta->insn.imm);
+}
+
+static int or_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu32_reg(nfp_prog, meta, ALU_OP_OR);
+}
+
+static int or_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu32_imm(nfp_prog, meta, ALU_OP_OR, !meta->insn.imm);
+}
+
+static int add_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu32_reg(nfp_prog, meta, ALU_OP_ADD);
+}
+
+static int add_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu32_imm(nfp_prog, meta, ALU_OP_ADD, !meta->insn.imm);
+}
+
+static int sub_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu32_reg(nfp_prog, meta, ALU_OP_SUB);
+}
+
+static int sub_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_alu32_imm(nfp_prog, meta, ALU_OP_SUB, !meta->insn.imm);
+}
+
+static int shl_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+
+	if (!insn->imm)
+		return 1; /* TODO: zero shift means indirect */
+
+	emit_shf(nfp_prog, reg_both(insn->dst_reg * 2),
+		 reg_none(), SHF_OP_NONE, reg_b(insn->dst_reg * 2),
+		 SHF_SC_L_SHF, insn->imm);
+	wrp_immed(nfp_prog, reg_both(insn->dst_reg * 2 + 1), 0);
+
+	return 0;
+}
+
+static int imm_ld8_part2(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	wrp_immed(nfp_prog, reg_both(nfp_meta_prev(meta)->insn.dst_reg * 2 + 1),
+		  meta->insn.imm);
+
+	return 0;
+}
+
+static int imm_ld8(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+
+	meta->double_cb = imm_ld8_part2;
+	wrp_immed(nfp_prog, reg_both(insn->dst_reg * 2), insn->imm);
+
+	return 0;
+}
+
+static int data_ld1(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return construct_data_ld(nfp_prog, meta->insn.imm, 1);
+}
+
+static int data_ld2(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return construct_data_ld(nfp_prog, meta->insn.imm, 2);
+}
+
+static int data_ld4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return construct_data_ld(nfp_prog, meta->insn.imm, 4);
+}
+
+static int data_ind_ld1(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return construct_data_ind_ld(nfp_prog, meta->insn.imm,
+				     meta->insn.src_reg * 2, true, 1);
+}
+
+static int data_ind_ld2(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return construct_data_ind_ld(nfp_prog, meta->insn.imm,
+				     meta->insn.src_reg * 2, true, 2);
+}
+
+static int data_ind_ld4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return construct_data_ind_ld(nfp_prog, meta->insn.imm,
+				     meta->insn.src_reg * 2, true, 4);
+}
+
+static int mem_ldx4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	if (meta->insn.off == offsetof(struct sk_buff, len))
+		emit_alu(nfp_prog, reg_both(meta->insn.dst_reg * 2),
+			 reg_none(), ALU_OP_NONE, NFP_BPF_ABI_LEN);
+	else
+		return -ENOTSUPP;
+
+	return 0;
+}
+
+static int jump(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	if (meta->insn.off < 0) /* TODO */
+		return -ENOTSUPP;
+	emit_br(nfp_prog, BR_UNC, meta->insn.off, 0);
+
+	return 0;
+}
+
+static int jeq_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+	u64 imm = insn->imm; /* sign extend */
+	u32 or1 = reg_a(insn->dst_reg * 2), or2 = reg_b(insn->dst_reg * 2 + 1);
+	u32 tmp_reg;
+
+	if (insn->off < 0) /* TODO */
+		return -ENOTSUPP;
+
+	if (imm & ~0U) {
+		tmp_reg = ur_load_imm_any(nfp_prog, imm & ~0U, imm_b(nfp_prog));
+		emit_alu(nfp_prog, imm_a(nfp_prog),
+			 reg_a(insn->dst_reg * 2), ALU_OP_XOR, tmp_reg);
+		or1 = imm_a(nfp_prog);
+	}
+
+	if (imm >> 32) {
+		tmp_reg = ur_load_imm_any(nfp_prog, imm >> 32, imm_b(nfp_prog));
+		emit_alu(nfp_prog, imm_b(nfp_prog),
+			 reg_a(insn->dst_reg * 2 + 1), ALU_OP_XOR, tmp_reg);
+		or2 = imm_b(nfp_prog);
+	}
+
+	emit_alu(nfp_prog, reg_none(), or1, ALU_OP_OR, or2);
+	emit_br(nfp_prog, BR_BEQ, insn->off, 0);
+
+	return 0;
+}
+
+static int jgt_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_cmp_imm(nfp_prog, meta, BR_BLO, false);
+}
+
+static int jge_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_cmp_imm(nfp_prog, meta, BR_BHS, true);
+}
+
+static int jset_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+	u64 imm = insn->imm; /* sign extend */
+	u32 tmp_reg;
+
+	if (insn->off < 0) /* TODO */
+		return -ENOTSUPP;
+
+	if (!imm) {
+		meta->skip = true;
+		return 0;
+	}
+
+	if (imm & ~0U) {
+		tmp_reg = ur_load_imm_any(nfp_prog, imm & ~0U, imm_b(nfp_prog));
+		emit_alu(nfp_prog, reg_none(),
+			 reg_a(insn->dst_reg * 2), ALU_OP_AND, tmp_reg);
+		emit_br(nfp_prog, BR_BNE, insn->off, 0);
+	}
+
+	if (imm >> 32) {
+		tmp_reg = ur_load_imm_any(nfp_prog, imm >> 32, imm_b(nfp_prog));
+		emit_alu(nfp_prog, reg_none(),
+			 reg_a(insn->dst_reg * 2 + 1), ALU_OP_AND, tmp_reg);
+		emit_br(nfp_prog, BR_BNE, insn->off, 0);
+	}
+
+	return 0;
+}
+
+static int jne_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+	u64 imm = insn->imm; /* sign extend */
+	u32 tmp_reg;
+
+	if (insn->off < 0) /* TODO */
+		return -ENOTSUPP;
+
+	if (!imm) {
+		emit_alu(nfp_prog, reg_none(), reg_a(insn->dst_reg * 2),
+			 ALU_OP_OR, reg_b(insn->dst_reg * 2 + 1));
+		emit_br(nfp_prog, BR_BNE, insn->off, 0);
+	}
+
+	tmp_reg = ur_load_imm_any(nfp_prog, imm & ~0U, imm_b(nfp_prog));
+	emit_alu(nfp_prog, reg_none(),
+		 reg_a(insn->dst_reg * 2), ALU_OP_XOR, tmp_reg);
+	emit_br(nfp_prog, BR_BNE, insn->off, 0);
+
+	tmp_reg = ur_load_imm_any(nfp_prog, imm >> 32, imm_b(nfp_prog));
+	emit_alu(nfp_prog, reg_none(),
+		 reg_a(insn->dst_reg * 2 + 1), ALU_OP_XOR, tmp_reg);
+	emit_br(nfp_prog, BR_BNE, insn->off, 0);
+
+	return 0;
+}
+
+static int jeq_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	const struct bpf_insn *insn = &meta->insn;
+
+	if (insn->off < 0) /* TODO */
+		return -ENOTSUPP;
+
+	emit_alu(nfp_prog, imm_a(nfp_prog), reg_a(insn->dst_reg * 2),
+		 ALU_OP_XOR, reg_b(insn->src_reg * 2));
+	emit_alu(nfp_prog, imm_b(nfp_prog), reg_a(insn->dst_reg * 2 + 1),
+		 ALU_OP_XOR, reg_b(insn->src_reg * 2 + 1));
+	emit_alu(nfp_prog, reg_none(),
+		 imm_a(nfp_prog), ALU_OP_OR, imm_b(nfp_prog));
+	emit_br(nfp_prog, BR_BEQ, insn->off, 0);
+
+	return 0;
+}
+
+static int jgt_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_cmp_reg(nfp_prog, meta, BR_BLO, false);
+}
+
+static int jge_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_cmp_reg(nfp_prog, meta, BR_BHS, true);
+}
+
+static int jset_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_test_reg(nfp_prog, meta, ALU_OP_AND, BR_BNE);
+}
+
+static int jne_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	return wrp_test_reg(nfp_prog, meta, ALU_OP_XOR, BR_BNE);
+}
+
+static int goto_out(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	wrp_br_special(nfp_prog, BR_UNC, OP_BR_GO_OUT);
+
+	return 0;
+}
+
+static const instr_cb_t instr_cb[256] = {
+	[BPF_ALU64 | BPF_MOV | BPF_X] =	mov_reg64,
+	[BPF_ALU64 | BPF_MOV | BPF_K] =	mov_imm64,
+	[BPF_ALU64 | BPF_XOR | BPF_X] =	xor_reg64,
+	[BPF_ALU64 | BPF_XOR | BPF_K] =	xor_imm64,
+	[BPF_ALU64 | BPF_AND | BPF_X] =	and_reg64,
+	[BPF_ALU64 | BPF_AND | BPF_K] =	and_imm64,
+	[BPF_ALU64 | BPF_OR | BPF_X] =	or_reg64,
+	[BPF_ALU64 | BPF_OR | BPF_K] =	or_imm64,
+	[BPF_ALU64 | BPF_ADD | BPF_X] =	add_reg64,
+	[BPF_ALU64 | BPF_ADD | BPF_K] =	add_imm64,
+	[BPF_ALU64 | BPF_SUB | BPF_X] =	sub_reg64,
+	[BPF_ALU64 | BPF_SUB | BPF_K] =	sub_imm64,
+	[BPF_ALU64 | BPF_LSH | BPF_K] =	shl_imm64,
+	[BPF_ALU64 | BPF_RSH | BPF_K] =	shr_imm64,
+	[BPF_ALU | BPF_MOV | BPF_X] =	mov_reg,
+	[BPF_ALU | BPF_MOV | BPF_K] =	mov_imm,
+	[BPF_ALU | BPF_XOR | BPF_X] =	xor_reg,
+	[BPF_ALU | BPF_XOR | BPF_K] =	xor_imm,
+	[BPF_ALU | BPF_AND | BPF_X] =	and_reg,
+	[BPF_ALU | BPF_AND | BPF_K] =	and_imm,
+	[BPF_ALU | BPF_OR | BPF_X] =	or_reg,
+	[BPF_ALU | BPF_OR | BPF_K] =	or_imm,
+	[BPF_ALU | BPF_ADD | BPF_X] =	add_reg,
+	[BPF_ALU | BPF_ADD | BPF_K] =	add_imm,
+	[BPF_ALU | BPF_SUB | BPF_X] =	sub_reg,
+	[BPF_ALU | BPF_SUB | BPF_K] =	sub_imm,
+	[BPF_ALU | BPF_LSH | BPF_K] =	shl_imm,
+	[BPF_LD | BPF_IMM | BPF_DW] =	imm_ld8,
+	[BPF_LD | BPF_ABS | BPF_B] =	data_ld1,
+	[BPF_LD | BPF_ABS | BPF_H] =	data_ld2,
+	[BPF_LD | BPF_ABS | BPF_W] =	data_ld4,
+	[BPF_LD | BPF_IND | BPF_B] =	data_ind_ld1,
+	[BPF_LD | BPF_IND | BPF_H] =	data_ind_ld2,
+	[BPF_LD | BPF_IND | BPF_W] =	data_ind_ld4,
+	[BPF_LDX | BPF_MEM | BPF_W] =	mem_ldx4,
+	[BPF_JMP | BPF_JA | BPF_K] =	jump,
+	[BPF_JMP | BPF_JEQ | BPF_K] =	jeq_imm,
+	[BPF_JMP | BPF_JGT | BPF_K] =	jgt_imm,
+	[BPF_JMP | BPF_JGE | BPF_K] =	jge_imm,
+	[BPF_JMP | BPF_JSET | BPF_K] =	jset_imm,
+	[BPF_JMP | BPF_JNE | BPF_K] =	jne_imm,
+	[BPF_JMP | BPF_JEQ | BPF_X] =	jeq_reg,
+	[BPF_JMP | BPF_JGT | BPF_X] =	jgt_reg,
+	[BPF_JMP | BPF_JGE | BPF_X] =	jge_reg,
+	[BPF_JMP | BPF_JSET | BPF_X] =	jset_reg,
+	[BPF_JMP | BPF_JNE | BPF_X] =	jne_reg,
+	[BPF_JMP | BPF_EXIT] =		goto_out,
+};
+
+/* --- Misc code --- */
+static void br_set_offset(u64 *instr, u16 offset)
+{
+	u16 addr_lo, addr_hi;
+
+	addr_lo = offset & (OP_BR_ADDR_LO >> __bf_shf(OP_BR_ADDR_LO));
+	addr_hi = offset != addr_lo;
+	*instr &= ~(OP_BR_ADDR_HI | OP_BR_ADDR_LO);
+	*instr |= FIELD_PREP(OP_BR_ADDR_HI, addr_hi);
+	*instr |= FIELD_PREP(OP_BR_ADDR_LO, addr_lo);
+}
+
+/* --- Assembler logic --- */
+static int nfp_fixup_branches(struct nfp_prog *nfp_prog)
+{
+	struct nfp_insn_meta *meta, *next;
+	u32 off, br_idx;
+	u32 idx;
+
+	nfp_for_each_insn_walk2(nfp_prog, meta, next) {
+		if (meta->skip)
+			continue;
+		if (BPF_CLASS(meta->insn.code) != BPF_JMP)
+			continue;
+
+		br_idx = nfp_prog_offset_to_index(nfp_prog, next->off) - 1;
+		if (!nfp_is_br(nfp_prog->prog[br_idx])) {
+			pr_err("Fixup found block not ending in branch %d %02x %016llx!!\n",
+			       br_idx, meta->insn.code, nfp_prog->prog[br_idx]);
+			return -ELOOP;
+		}
+		/* Leave special branches for later */
+		if (FIELD_GET(OP_BR_SPECIAL, nfp_prog->prog[br_idx]))
+			continue;
+
+		/* Find the target offset in assembler realm */
+		off = meta->insn.off;
+		if (!off) {
+			pr_err("Fixup found zero offset!!\n");
+			return -ELOOP;
+		}
+
+		while (off && nfp_meta_has_next(nfp_prog, next)) {
+			next = nfp_meta_next(next);
+			off--;
+		}
+		if (off) {
+			pr_err("Fixup found too large jump!! %d\n", off);
+			return -ELOOP;
+		}
+
+		if (next->skip) {
+			pr_err("Branch landing on removed instruction!!\n");
+			return -ELOOP;
+		}
+
+		for (idx = nfp_prog_offset_to_index(nfp_prog, meta->off);
+		     idx <= br_idx; idx++) {
+			if (!nfp_is_br(nfp_prog->prog[idx]))
+				continue;
+			br_set_offset(&nfp_prog->prog[idx], next->off);
+		}
+	}
+
+	/* Fixup 'goto out's separately, they can be scattered around */
+	for (br_idx = 0; br_idx < nfp_prog->prog_len; br_idx++) {
+		enum br_special special;
+
+		if ((nfp_prog->prog[br_idx] & OP_BR_BASE_MASK) != OP_BR_BASE)
+			continue;
+
+		special = FIELD_GET(OP_BR_SPECIAL, nfp_prog->prog[br_idx]);
+		switch (special) {
+		case OP_BR_NORMAL:
+			break;
+		case OP_BR_GO_OUT:
+			br_set_offset(&nfp_prog->prog[br_idx],
+				      nfp_prog->tgt_out);
+			break;
+		case OP_BR_GO_ABORT:
+			br_set_offset(&nfp_prog->prog[br_idx],
+				      nfp_prog->tgt_abort);
+			break;
+		}
+
+		nfp_prog->prog[br_idx] &= ~OP_BR_SPECIAL;
+	}
+
+	return 0;
+}
+
+static void nfp_intro(struct nfp_prog *nfp_prog)
+{
+	emit_alu(nfp_prog, pkt_reg(nfp_prog),
+		 reg_none(), ALU_OP_NONE, NFP_BPF_ABI_PKT);
+}
+
+static void nfp_outro_tc_legacy(struct nfp_prog *nfp_prog)
+{
+	const u8 act2code[] = {
+		[NN_ACT_TC_DROP]  = 0x22,
+	};
+	/* Target for aborts */
+	nfp_prog->tgt_abort = nfp_prog_current_offset(nfp_prog);
+	wrp_immed(nfp_prog, reg_both(0), 0);
+
+	/* Target for normal exits */
+	nfp_prog->tgt_out = nfp_prog_current_offset(nfp_prog);
+	/* Legacy TC mode:
+	 *   0        0x11 -> pass,  count as stat0
+	 *  -1  drop  0x22 -> drop,  count as stat1
+	 *     redir  0x24 -> redir, count as stat1
+	 *  ife mark  0x21 -> pass,  count as stat1
+	 *  ife + tx  0x24 -> redir, count as stat1
+	 */
+	emit_br_byte_neq(nfp_prog, reg_b(0), 0xff, 0, nfp_prog->tgt_done, 2);
+	emit_alu(nfp_prog, reg_a(0),
+		 reg_none(), ALU_OP_NONE, NFP_BPF_ABI_FLAGS);
+	emit_ld_field(nfp_prog, reg_a(0), 0xc, reg_imm(0x11), SHF_SC_L_SHF, 16);
+
+	emit_br(nfp_prog, BR_UNC, nfp_prog->tgt_done, 1);
+	emit_ld_field(nfp_prog, reg_a(0), 0xc, reg_imm(act2code[nfp_prog->act]),
+		      SHF_SC_L_SHF, 16);
+}
+
+static void nfp_outro(struct nfp_prog *nfp_prog)
+{
+	switch (nfp_prog->act) {
+	case NN_ACT_TC_DROP:
+		nfp_outro_tc_legacy(nfp_prog);
+		break;
+	}
+}
+
+static int nfp_translate(struct nfp_prog *nfp_prog)
+{
+	struct nfp_insn_meta *meta;
+	int err;
+
+	nfp_intro(nfp_prog);
+	if (nfp_prog->error)
+		return nfp_prog->error;
+
+	list_for_each_entry(meta, &nfp_prog->insns, l) {
+		instr_cb_t cb = instr_cb[meta->insn.code];
+
+		meta->off = nfp_prog_current_offset(nfp_prog);
+
+		if (meta->skip) {
+			nfp_prog->n_translated++;
+			continue;
+		}
+
+		if (nfp_meta_has_prev(nfp_prog, meta) &&
+		    nfp_meta_prev(meta)->double_cb)
+			cb = nfp_meta_prev(meta)->double_cb;
+		if (!cb)
+			return -ENOENT;
+		err = cb(nfp_prog, meta);
+		if (err)
+			return err;
+
+		nfp_prog->n_translated++;
+	}
+
+	nfp_outro(nfp_prog);
+	if (nfp_prog->error)
+		return nfp_prog->error;
+
+	return nfp_fixup_branches(nfp_prog);
+}
+
+static int
+nfp_prog_prepare(struct nfp_prog *nfp_prog, const struct bpf_insn *prog,
+		 unsigned int cnt)
+{
+	unsigned int i;
+
+	for (i = 0; i < cnt; i++) {
+		struct nfp_insn_meta *meta;
+
+		meta = kzalloc(sizeof(*meta), GFP_KERNEL);
+		if (!meta)
+			return -ENOMEM;
+
+		meta->insn = prog[i];
+		meta->n = i;
+
+		list_add_tail(&meta->l, &nfp_prog->insns);
+	}
+
+	return 0;
+}
+
+/* --- Optimizations --- */
+static void nfp_bpf_opt_reg_init(struct nfp_prog *nfp_prog)
+{
+	struct nfp_insn_meta *meta;
+
+	list_for_each_entry(meta, &nfp_prog->insns, l) {
+		struct bpf_insn insn = meta->insn;
+
+		/* Programs converted from cBPF start with register xoring */
+		if (insn.code == (BPF_ALU64 | BPF_XOR | BPF_X) &&
+		    insn.src_reg == insn.dst_reg)
+			continue;
+
+		/* Programs start with R6 = R1 but we ignore the skb pointer */
+		if (insn.code == (BPF_ALU64 | BPF_MOV | BPF_X) &&
+		    insn.src_reg == 1 && insn.dst_reg == 6)
+			meta->skip = true;
+
+		/* Return as soon as something doesn't match */
+		if (!meta->skip)
+			return;
+	}
+}
+
+/* Try to rename registers so that program uses only low ones */
+static int nfp_bpf_opt_reg_rename(struct nfp_prog *nfp_prog)
+{
+	bool reg_used[MAX_BPF_REG] = {};
+	u8 tgt_reg[MAX_BPF_REG] = {};
+	struct nfp_insn_meta *meta;
+	unsigned int i, j;
+
+	list_for_each_entry(meta, &nfp_prog->insns, l) {
+		if (meta->skip)
+			continue;
+
+		reg_used[meta->insn.src_reg] = true;
+		reg_used[meta->insn.dst_reg] = true;
+	}
+
+	if (reg_used[BPF_REG_10]) {
+		pr_err("Detected use of stack ptr\n");
+		return -EINVAL;
+	}
+
+	for (i = 0, j = 0; i < ARRAY_SIZE(tgt_reg); i++) {
+		if (!reg_used[i])
+			continue;
+
+		tgt_reg[i] = j++;
+	}
+	nfp_prog->num_regs = j;
+
+	list_for_each_entry(meta, &nfp_prog->insns, l) {
+		meta->insn.src_reg = tgt_reg[meta->insn.src_reg];
+		meta->insn.dst_reg = tgt_reg[meta->insn.dst_reg];
+	}
+
+	return 0;
+}
+
+/* Remove masking after load since our load guarantees this is not needed */
+static void nfp_bpf_opt_ld_mask(struct nfp_prog *nfp_prog)
+{
+	struct nfp_insn_meta *meta1, *meta2;
+	const s32 exp_mask[] = {
+		[BPF_B] = 0x000000ffU,
+		[BPF_H] = 0x0000ffffU,
+		[BPF_W] = 0xffffffffU,
+	};
+
+	nfp_for_each_insn_walk2(nfp_prog, meta1, meta2) {
+		struct bpf_insn insn, next;
+
+		insn = meta1->insn;
+		next = meta2->insn;
+
+		if (BPF_CLASS(insn.code) != BPF_LD)
+			continue;
+		if (BPF_MODE(insn.code) != BPF_ABS &&
+		    BPF_MODE(insn.code) != BPF_IND)
+			continue;
+
+		if (next.code != (BPF_ALU64 | BPF_AND | BPF_K))
+			continue;
+
+		if (!exp_mask[BPF_SIZE(insn.code)])
+			continue;
+		if (exp_mask[BPF_SIZE(insn.code)] != next.imm)
+			continue;
+
+		if (next.src_reg || next.dst_reg)
+			continue;
+
+		meta2->skip = true;
+	}
+}
+
+static void nfp_bpf_opt_ld_shift(struct nfp_prog *nfp_prog)
+{
+	struct nfp_insn_meta *meta1, *meta2, *meta3;
+
+	nfp_for_each_insn_walk3(nfp_prog, meta1, meta2, meta3) {
+		struct bpf_insn insn, next1, next2;
+
+		insn = meta1->insn;
+		next1 = meta2->insn;
+		next2 = meta3->insn;
+
+		if (BPF_CLASS(insn.code) != BPF_LD)
+			continue;
+		if (BPF_MODE(insn.code) != BPF_ABS &&
+		    BPF_MODE(insn.code) != BPF_IND)
+			continue;
+		if (BPF_SIZE(insn.code) != BPF_W)
+			continue;
+
+		if (!(next1.code == (BPF_LSH | BPF_K | BPF_ALU64) &&
+		      next2.code == (BPF_RSH | BPF_K | BPF_ALU64)) &&
+		    !(next1.code == (BPF_RSH | BPF_K | BPF_ALU64) &&
+		      next2.code == (BPF_LSH | BPF_K | BPF_ALU64)))
+			continue;
+
+		if (next1.src_reg || next1.dst_reg ||
+		    next2.src_reg || next2.dst_reg)
+			continue;
+
+		if (next1.imm != 0x20 || next2.imm != 0x20)
+			continue;
+
+		meta2->skip = true;
+		meta3->skip = true;
+	}
+}
+
+static int nfp_bpf_optimize(struct nfp_prog *nfp_prog)
+{
+	int ret;
+
+	nfp_bpf_opt_reg_init(nfp_prog);
+
+	ret = nfp_bpf_opt_reg_rename(nfp_prog);
+	if (ret)
+		return ret;
+
+	nfp_bpf_opt_ld_mask(nfp_prog);
+	nfp_bpf_opt_ld_shift(nfp_prog);
+
+	return 0;
+}
+
+/**
+ * nfp_bpf_jit() - translate BPF code into NFP assembly
+ * @filter:	kernel BPF filter struct
+ * @prog_mem:	memory to store assembler instructions
+ * @act:	action attached to this eBPF program
+ * @prog_start:	offset of the first instruction when loaded
+ * @prog_done:	where to jump on exit
+ * @prog_sz:	size of @prog_mem in instructions
+ * @res:	achieved parameters of translation results
+ */
+int
+nfp_bpf_jit(struct bpf_prog *filter, void *prog_mem,
+	    enum nfp_bpf_action_type act,
+	    unsigned int prog_start, unsigned int prog_done,
+	    unsigned int prog_sz, struct nfp_bpf_result *res)
+{
+	struct nfp_prog *nfp_prog;
+	int ret;
+
+	nfp_prog = kzalloc(sizeof(*nfp_prog), GFP_KERNEL);
+	if (!nfp_prog)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&nfp_prog->insns);
+	nfp_prog->act = act;
+	nfp_prog->start_off = prog_start;
+	nfp_prog->tgt_done = prog_done;
+
+	ret = nfp_prog_prepare(nfp_prog, filter->insnsi, filter->len);
+	if (ret)
+		goto out;
+
+	ret = nfp_prog_verify(nfp_prog, filter);
+	if (ret)
+		return ret;
+
+	ret = nfp_bpf_optimize(nfp_prog);
+	if (ret)
+		goto out;
+
+	if (nfp_prog->num_regs <= 7)
+		nfp_prog->regs_per_thread = 16;
+	else
+		nfp_prog->regs_per_thread = 32;
+
+	nfp_prog->prog = prog_mem;
+	nfp_prog->__prog_alloc_len = prog_sz;
+
+	ret = nfp_translate(nfp_prog);
+	if (ret) {
+		pr_err("Translation failed with error %d (translated: %u)\n",
+		       ret, nfp_prog->n_translated);
+		ret = -EINVAL;
+	}
+
+	res->n_instr = nfp_prog->prog_len;
+	res->dense_mode = nfp_prog->num_regs <= 7;
+out:
+	nfp_prog_free(nfp_prog);
+
+	return ret;
+}
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_bpf_verifier.c b/drivers/net/ethernet/netronome/nfp/nfp_bpf_verifier.c
new file mode 100644
index 000000000000..15c460964810
--- /dev/null
+++ b/drivers/net/ethernet/netronome/nfp/nfp_bpf_verifier.c
@@ -0,0 +1,151 @@
+/*
+ * 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.
+ */
+
+#define pr_fmt(fmt)	"NFP net bpf: " fmt
+
+#include <linux/bpf.h>
+#include <linux/bpf_verifier.h>
+#include <linux/kernel.h>
+#include <linux/pkt_cls.h>
+
+#include "nfp_bpf.h"
+
+/* Analyzer/verifier definitions */
+struct nfp_bpf_analyzer_priv {
+	struct nfp_prog *prog;
+	struct nfp_insn_meta *meta;
+};
+
+static struct nfp_insn_meta *
+nfp_bpf_goto_meta(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
+		  unsigned int insn_idx, unsigned int n_insns)
+{
+	unsigned int forward, backward, i;
+
+	backward = meta->n - insn_idx;
+	forward = insn_idx - meta->n;
+
+	if (min(forward, backward) > n_insns - insn_idx - 1) {
+		backward = n_insns - insn_idx - 1;
+		meta = nfp_prog_last_meta(nfp_prog);
+	}
+	if (min(forward, backward) > insn_idx && backward > insn_idx) {
+		forward = insn_idx;
+		meta = nfp_prog_first_meta(nfp_prog);
+	}
+
+	if (forward < backward)
+		for (i = 0; i < forward; i++)
+			meta = nfp_meta_next(meta);
+	else
+		for (i = 0; i < backward; i++)
+			meta = nfp_meta_prev(meta);
+
+	return meta;
+}
+
+static int
+nfp_bpf_check_exit(struct nfp_prog *nfp_prog,
+		   const struct bpf_verifier_env *env)
+{
+	const struct bpf_reg_state *reg0 = &env->cur_state.regs[0];
+
+	if (reg0->type != CONST_IMM) {
+		pr_info("unsupported exit state: %d, imm: %llx\n",
+			reg0->type, reg0->imm);
+		return -EINVAL;
+	}
+
+	if (reg0->imm != 0 && (reg0->imm & ~0U) != ~0U) {
+		pr_info("unsupported exit state: %d, imm: %llx\n",
+			reg0->type, reg0->imm);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int
+nfp_bpf_check_ctx_ptr(struct nfp_prog *nfp_prog,
+		      const struct bpf_verifier_env *env, u8 reg)
+{
+	if (env->cur_state.regs[reg].type != PTR_TO_CTX)
+		return -EINVAL;
+
+	return 0;
+}
+
+static int
+nfp_verify_insn(struct bpf_verifier_env *env, int insn_idx, int prev_insn_idx)
+{
+	struct nfp_bpf_analyzer_priv *priv = env->analyzer_priv;
+	struct nfp_insn_meta *meta = priv->meta;
+
+	meta = nfp_bpf_goto_meta(priv->prog, meta, insn_idx, env->prog->len);
+	priv->meta = meta;
+
+	if (meta->insn.code == (BPF_JMP | BPF_EXIT))
+		return nfp_bpf_check_exit(priv->prog, env);
+
+	if ((meta->insn.code & ~BPF_SIZE_MASK) == (BPF_LDX | BPF_MEM))
+		return nfp_bpf_check_ctx_ptr(priv->prog, env,
+					     meta->insn.src_reg);
+	if ((meta->insn.code & ~BPF_SIZE_MASK) == (BPF_STX | BPF_MEM))
+		return nfp_bpf_check_ctx_ptr(priv->prog, env,
+					     meta->insn.dst_reg);
+
+	return 0;
+}
+
+static const struct bpf_ext_analyzer_ops nfp_bpf_analyzer_ops = {
+	.insn_hook = nfp_verify_insn,
+};
+
+int nfp_prog_verify(struct nfp_prog *nfp_prog, struct bpf_prog *prog)
+{
+	struct nfp_bpf_analyzer_priv *priv;
+	int ret;
+
+	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+	if (!priv)
+		return -ENOMEM;
+
+	priv->prog = nfp_prog;
+	priv->meta = nfp_prog_first_meta(nfp_prog);
+
+	ret = bpf_analyzer(prog, &nfp_bpf_analyzer_ops, priv);
+
+	kfree(priv);
+
+	return ret;
+}
-- 
1.9.1

[PATCHv5 net-next 09/15] nfp: bpf: add hardware bpf offload

[Jakub Kicinski]

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>
---
 drivers/net/ethernet/netronome/nfp/Makefile        |   1 +
 drivers/net/ethernet/netronome/nfp/nfp_net.h       |  26 ++-
 .../net/ethernet/netronome/nfp/nfp_net_common.c    |  40 +++-
 drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h  |  44 ++++-
 .../net/ethernet/netronome/nfp/nfp_net_offload.c   | 220 +++++++++++++++++++++
 5 files changed, 324 insertions(+), 7 deletions(-)
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_net_offload.c

diff --git a/drivers/net/ethernet/netronome/nfp/Makefile b/drivers/net/ethernet/netronome/nfp/Makefile
index 5f12689bf523..0efb2ba9a558 100644
--- a/drivers/net/ethernet/netronome/nfp/Makefile
+++ b/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)
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net.h b/drivers/net/ethernet/netronome/nfp/nfp_net.h
index 690635660195..ea6f5e667f27 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net.h
+++ b/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_ */
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index 252e4924de0f..51978dfe883b 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/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);
 
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h b/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h
index ad6c4e31cedd..7aa11f3c5ef0 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h
+++ b/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
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c b/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
new file mode 100644
index 000000000000..313988cd3a43
--- /dev/null
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
@@ -0,0 +1,220 @@
+/*
+ * 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;
+	}
+}
-- 
1.9.1

[PATCHv5 net-next 10/15] net: cls_bpf: allow offloaded filters to update stats

[Jakub Kicinski]

Call into offloaded filters to update stats.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
---
 include/net/pkt_cls.h |  1 +
 net/sched/cls_bpf.c   | 11 +++++++++++
 2 files changed, 12 insertions(+)

diff --git a/include/net/pkt_cls.h b/include/net/pkt_cls.h
index 57af9f3032ff..5ccaa4be7d96 100644
--- a/include/net/pkt_cls.h
+++ b/include/net/pkt_cls.h
@@ -490,6 +490,7 @@ enum tc_clsbpf_command {
 	TC_CLSBPF_ADD,
 	TC_CLSBPF_REPLACE,
 	TC_CLSBPF_DESTROY,
+	TC_CLSBPF_STATS,
 };
 
 struct tc_cls_bpf_offload {
diff --git a/net/sched/cls_bpf.c b/net/sched/cls_bpf.c
index 18f9869cd4da..9b29e0673346 100644
--- a/net/sched/cls_bpf.c
+++ b/net/sched/cls_bpf.c
@@ -220,6 +220,15 @@ static void cls_bpf_stop_offload(struct tcf_proto *tp,
 	prog->offloaded = false;
 }
 
+static void cls_bpf_offload_update_stats(struct tcf_proto *tp,
+					 struct cls_bpf_prog *prog)
+{
+	if (!prog->offloaded)
+		return;
+
+	cls_bpf_offload_cmd(tp, prog, TC_CLSBPF_STATS);
+}
+
 static int cls_bpf_init(struct tcf_proto *tp)
 {
 	struct cls_bpf_head *head;
@@ -575,6 +584,8 @@ static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
 
 	tm->tcm_handle = prog->handle;
 
+	cls_bpf_offload_update_stats(tp, prog);
+
 	nest = nla_nest_start(skb, TCA_OPTIONS);
 	if (nest == NULL)
 		goto nla_put_failure;
-- 
1.9.1

[PATCHv5 net-next 11/15] nfp: bpf: allow offloaded filters to update stats

[Jakub Kicinski]

Periodically poll stats and call into offloaded actions
to update them.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
v3:
 - add missing hunk with ethtool stats.
---
 drivers/net/ethernet/netronome/nfp/nfp_net.h       | 19 +++++++
 .../net/ethernet/netronome/nfp/nfp_net_common.c    |  3 ++
 .../net/ethernet/netronome/nfp/nfp_net_ethtool.c   | 12 +++++
 .../net/ethernet/netronome/nfp/nfp_net_offload.c   | 63 ++++++++++++++++++++++
 4 files changed, 97 insertions(+)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net.h b/drivers/net/ethernet/netronome/nfp/nfp_net.h
index ea6f5e667f27..13c6a9001b4d 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net.h
@@ -62,6 +62,9 @@
 /* Max time to wait for NFP to respond on updates (in seconds) */
 #define NFP_NET_POLL_TIMEOUT	5
 
+/* Interval for reading offloaded filter stats */
+#define NFP_NET_STAT_POLL_IVL	msecs_to_jiffies(100)
+
 /* Bar allocation */
 #define NFP_NET_CTRL_BAR	0
 #define NFP_NET_Q0_BAR		2
@@ -405,6 +408,11 @@ static inline bool nfp_net_fw_ver_eq(struct nfp_net_fw_version *fw_ver,
 	       fw_ver->minor == minor;
 }
 
+struct nfp_stat_pair {
+	u64 pkts;
+	u64 bytes;
+};
+
 /**
  * struct nfp_net - NFP network device structure
  * @pdev:               Backpointer to PCI device
@@ -428,6 +436,11 @@ static inline bool nfp_net_fw_ver_eq(struct nfp_net_fw_version *fw_ver,
  * @rss_cfg:            RSS configuration
  * @rss_key:            RSS secret key
  * @rss_itbl:           RSS indirection table
+ * @rx_filter:		Filter offload statistics - dropped packets/bytes
+ * @rx_filter_prev:	Filter offload statistics - values from previous update
+ * @rx_filter_change:	Jiffies when statistics last changed
+ * @rx_filter_stats_timer:  Timer for polling filter offload statistics
+ * @rx_filter_lock:	Lock protecting timer state changes (teardown)
  * @max_tx_rings:       Maximum number of TX rings supported by the Firmware
  * @max_rx_rings:       Maximum number of RX rings supported by the Firmware
  * @num_tx_rings:       Currently configured number of TX rings
@@ -504,6 +517,11 @@ struct nfp_net {
 	u8 rss_key[NFP_NET_CFG_RSS_KEY_SZ];
 	u8 rss_itbl[NFP_NET_CFG_RSS_ITBL_SZ];
 
+	struct nfp_stat_pair rx_filter, rx_filter_prev;
+	unsigned long rx_filter_change;
+	struct timer_list rx_filter_stats_timer;
+	spinlock_t rx_filter_lock;
+
 	int max_tx_rings;
 	int max_rx_rings;
 
@@ -775,6 +793,7 @@ static inline void nfp_net_debugfs_adapter_del(struct nfp_net *nn)
 }
 #endif /* CONFIG_NFP_NET_DEBUG */
 
+void nfp_net_filter_stats_timer(unsigned long data);
 int
 nfp_net_bpf_offload(struct nfp_net *nn, u32 handle, __be16 proto,
 		    struct tc_cls_bpf_offload *cls_bpf);
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index 51978dfe883b..f091eb758ca2 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -2703,10 +2703,13 @@ struct nfp_net *nfp_net_netdev_alloc(struct pci_dev *pdev,
 	nn->rxd_cnt = NFP_NET_RX_DESCS_DEFAULT;
 
 	spin_lock_init(&nn->reconfig_lock);
+	spin_lock_init(&nn->rx_filter_lock);
 	spin_lock_init(&nn->link_status_lock);
 
 	setup_timer(&nn->reconfig_timer,
 		    nfp_net_reconfig_timer, (unsigned long)nn);
+	setup_timer(&nn->rx_filter_stats_timer,
+		    nfp_net_filter_stats_timer, (unsigned long)nn);
 
 	return nn;
 }
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c b/drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c
index 4c9897220969..3418f2277e9d 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c
@@ -106,6 +106,18 @@ static const struct _nfp_net_et_stats nfp_net_et_stats[] = {
 	{"dev_tx_pkts", NN_ET_DEV_STAT(NFP_NET_CFG_STATS_TX_FRAMES)},
 	{"dev_tx_mc_pkts", NN_ET_DEV_STAT(NFP_NET_CFG_STATS_TX_MC_FRAMES)},
 	{"dev_tx_bc_pkts", NN_ET_DEV_STAT(NFP_NET_CFG_STATS_TX_BC_FRAMES)},
+
+	{"bpf_pass_pkts", NN_ET_DEV_STAT(NFP_NET_CFG_STATS_APP0_FRAMES)},
+	{"bpf_pass_bytes", NN_ET_DEV_STAT(NFP_NET_CFG_STATS_APP0_BYTES)},
+	/* see comments in outro functions in nfp_bpf_jit.c to find out
+	 * how different BPF modes use app-specific counters
+	 */
+	{"bpf_app1_pkts", NN_ET_DEV_STAT(NFP_NET_CFG_STATS_APP1_FRAMES)},
+	{"bpf_app1_bytes", NN_ET_DEV_STAT(NFP_NET_CFG_STATS_APP1_BYTES)},
+	{"bpf_app2_pkts", NN_ET_DEV_STAT(NFP_NET_CFG_STATS_APP2_FRAMES)},
+	{"bpf_app2_bytes", NN_ET_DEV_STAT(NFP_NET_CFG_STATS_APP2_BYTES)},
+	{"bpf_app3_pkts", NN_ET_DEV_STAT(NFP_NET_CFG_STATS_APP3_FRAMES)},
+	{"bpf_app3_bytes", NN_ET_DEV_STAT(NFP_NET_CFG_STATS_APP3_BYTES)},
 };
 
 #define NN_ET_GLOBAL_STATS_LEN ARRAY_SIZE(nfp_net_et_stats)
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c b/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
index 313988cd3a43..0537a53e2174 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
@@ -51,6 +51,60 @@
 #include "nfp_net_ctrl.h"
 #include "nfp_net.h"
 
+void nfp_net_filter_stats_timer(unsigned long data)
+{
+	struct nfp_net *nn = (void *)data;
+	struct nfp_stat_pair latest;
+
+	spin_lock_bh(&nn->rx_filter_lock);
+
+	if (nn->ctrl & NFP_NET_CFG_CTRL_BPF)
+		mod_timer(&nn->rx_filter_stats_timer,
+			  jiffies + NFP_NET_STAT_POLL_IVL);
+
+	spin_unlock_bh(&nn->rx_filter_lock);
+
+	latest.pkts = nn_readq(nn, NFP_NET_CFG_STATS_APP1_FRAMES);
+	latest.bytes = nn_readq(nn, NFP_NET_CFG_STATS_APP1_BYTES);
+
+	if (latest.pkts != nn->rx_filter.pkts)
+		nn->rx_filter_change = jiffies;
+
+	nn->rx_filter = latest;
+}
+
+static void nfp_net_bpf_stats_reset(struct nfp_net *nn)
+{
+	nn->rx_filter.pkts = nn_readq(nn, NFP_NET_CFG_STATS_APP1_FRAMES);
+	nn->rx_filter.bytes = nn_readq(nn, NFP_NET_CFG_STATS_APP1_BYTES);
+	nn->rx_filter_prev = nn->rx_filter;
+	nn->rx_filter_change = jiffies;
+}
+
+static int
+nfp_net_bpf_stats_update(struct nfp_net *nn, struct tc_cls_bpf_offload *cls_bpf)
+{
+	struct tc_action *a;
+	LIST_HEAD(actions);
+	u64 bytes, pkts;
+
+	pkts = nn->rx_filter.pkts - nn->rx_filter_prev.pkts;
+	bytes = nn->rx_filter.bytes - nn->rx_filter_prev.bytes;
+	bytes -= pkts * ETH_HLEN;
+
+	nn->rx_filter_prev = nn->rx_filter;
+
+	preempt_disable();
+
+	tcf_exts_to_list(cls_bpf->exts, &actions);
+	list_for_each_entry(a, &actions, list)
+		tcf_action_stats_update(a, bytes, pkts, nn->rx_filter_change);
+
+	preempt_enable();
+
+	return 0;
+}
+
 static int
 nfp_net_bpf_get_act(struct nfp_net *nn, struct tc_cls_bpf_offload *cls_bpf)
 {
@@ -147,6 +201,9 @@ nfp_net_bpf_load_and_start(struct nfp_net *nn, u32 tc_flags,
 		nn_err(nn, "FW command error while enabling BPF: %d\n", err);
 
 	dma_free_coherent(&nn->pdev->dev, code_sz, code, dma_addr);
+
+	nfp_net_bpf_stats_reset(nn);
+	mod_timer(&nn->rx_filter_stats_timer, jiffies + NFP_NET_STAT_POLL_IVL);
 }
 
 static int nfp_net_bpf_stop(struct nfp_net *nn)
@@ -154,9 +211,12 @@ static int nfp_net_bpf_stop(struct nfp_net *nn)
 	if (!(nn->ctrl & NFP_NET_CFG_CTRL_BPF))
 		return 0;
 
+	spin_lock_bh(&nn->rx_filter_lock);
 	nn->ctrl &= ~NFP_NET_CFG_CTRL_BPF;
+	spin_unlock_bh(&nn->rx_filter_lock);
 	nn_writel(nn, NFP_NET_CFG_CTRL, nn->ctrl);
 
+	del_timer_sync(&nn->rx_filter_stats_timer);
 	nn->bpf_offload_skip_sw = 0;
 
 	return nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
@@ -214,6 +274,9 @@ nfp_net_bpf_offload(struct nfp_net *nn, u32 handle, __be16 proto,
 	case TC_CLSBPF_DESTROY:
 		return nfp_net_bpf_stop(nn);
 
+	case TC_CLSBPF_STATS:
+		return nfp_net_bpf_stats_update(nn, cls_bpf);
+
 	default:
 		return -ENOTSUPP;
 	}
-- 
1.9.1

[PATCHv5 net-next 12/15] nfp: bpf: add packet marking support

[Jakub Kicinski]

Add missing ABI defines and eBPF instructions to allow
mark to be passed on and extend prepend parsing on the
RX path to pick it up from packet metadata.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
v3:
 - change metadata format.
---
 drivers/net/ethernet/netronome/nfp/nfp_bpf.h       |  2 +
 drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c   | 19 +++++
 drivers/net/ethernet/netronome/nfp/nfp_net.h       |  2 +
 .../net/ethernet/netronome/nfp/nfp_net_common.c    | 91 +++++++++++++++++-----
 drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h  |  7 ++
 .../net/ethernet/netronome/nfp/nfp_netvf_main.c    |  2 +-
 6 files changed, 101 insertions(+), 22 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_bpf.h b/drivers/net/ethernet/netronome/nfp/nfp_bpf.h
index af43e058be97..d550fbc4768a 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_bpf.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_bpf.h
@@ -91,6 +91,8 @@ enum nfp_bpf_reg_type {
 #define imm_both(np)	reg_both((np)->regs_per_thread - STATIC_REG_IMM)
 
 #define NFP_BPF_ABI_FLAGS	reg_nnr(0)
+#define   NFP_BPF_ABI_FLAG_MARK	1
+#define NFP_BPF_ABI_MARK	reg_nnr(1)
 #define NFP_BPF_ABI_PKT		reg_nnr(2)
 #define NFP_BPF_ABI_LEN		reg_nnr(3)
 
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c b/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
index dfcf162ccbb8..42a8afb67fc8 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
@@ -674,6 +674,16 @@ static int construct_data_ld(struct nfp_prog *nfp_prog, u16 offset, u8 size)
 	return construct_data_ind_ld(nfp_prog, offset, 0, false, size);
 }
 
+static int wrp_set_mark(struct nfp_prog *nfp_prog, u8 src)
+{
+	emit_alu(nfp_prog, NFP_BPF_ABI_MARK,
+		 reg_none(), ALU_OP_NONE, reg_b(src));
+	emit_alu(nfp_prog, NFP_BPF_ABI_FLAGS,
+		 NFP_BPF_ABI_FLAGS, ALU_OP_OR, reg_imm(NFP_BPF_ABI_FLAG_MARK));
+
+	return 0;
+}
+
 static void
 wrp_alu_imm(struct nfp_prog *nfp_prog, u8 dst, enum alu_op alu_op, u32 imm)
 {
@@ -1117,6 +1127,14 @@ static int mem_ldx4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 	return 0;
 }
 
+static int mem_stx4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	if (meta->insn.off == offsetof(struct sk_buff, mark))
+		return wrp_set_mark(nfp_prog, meta->insn.src_reg * 2);
+
+	return -ENOTSUPP;
+}
+
 static int jump(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 {
 	if (meta->insn.off < 0) /* TODO */
@@ -1306,6 +1324,7 @@ static const instr_cb_t instr_cb[256] = {
 	[BPF_LD | BPF_IND | BPF_H] =	data_ind_ld2,
 	[BPF_LD | BPF_IND | BPF_W] =	data_ind_ld4,
 	[BPF_LDX | BPF_MEM | BPF_W] =	mem_ldx4,
+	[BPF_STX | BPF_MEM | BPF_W] =	mem_stx4,
 	[BPF_JMP | BPF_JA | BPF_K] =	jump,
 	[BPF_JMP | BPF_JEQ | BPF_K] =	jeq_imm,
 	[BPF_JMP | BPF_JGT | BPF_K] =	jgt_imm,
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net.h b/drivers/net/ethernet/netronome/nfp/nfp_net.h
index 13c6a9001b4d..ed824e11a1e3 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net.h
@@ -269,6 +269,8 @@ struct nfp_net_rx_desc {
 	};
 };
 
+#define NFP_NET_META_FIELD_MASK GENMASK(NFP_NET_META_FIELD_SIZE - 1, 0)
+
 struct nfp_net_rx_hash {
 	__be32 hash_type;
 	__be32 hash;
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index f091eb758ca2..415691edcaa5 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -1293,38 +1293,72 @@ static void nfp_net_rx_csum(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
 	}
 }
 
-/**
- * nfp_net_set_hash() - Set SKB hash data
- * @netdev: adapter's net_device structure
- * @skb:   SKB to set the hash data on
- * @rxd:   RX descriptor
- *
- * The RSS hash and hash-type are pre-pended to the packet data.
- * Extract and decode it and set the skb fields.
- */
 static void nfp_net_set_hash(struct net_device *netdev, struct sk_buff *skb,
-			     struct nfp_net_rx_desc *rxd)
+			     unsigned int type, __be32 *hash)
 {
-	struct nfp_net_rx_hash *rx_hash;
-
-	if (!(rxd->rxd.flags & PCIE_DESC_RX_RSS) ||
-	    !(netdev->features & NETIF_F_RXHASH))
+	if (!(netdev->features & NETIF_F_RXHASH))
 		return;
 
-	rx_hash = (struct nfp_net_rx_hash *)(skb->data - sizeof(*rx_hash));
-
-	switch (be32_to_cpu(rx_hash->hash_type)) {
+	switch (type) {
 	case NFP_NET_RSS_IPV4:
 	case NFP_NET_RSS_IPV6:
 	case NFP_NET_RSS_IPV6_EX:
-		skb_set_hash(skb, be32_to_cpu(rx_hash->hash), PKT_HASH_TYPE_L3);
+		skb_set_hash(skb, get_unaligned_be32(hash), PKT_HASH_TYPE_L3);
 		break;
 	default:
-		skb_set_hash(skb, be32_to_cpu(rx_hash->hash), PKT_HASH_TYPE_L4);
+		skb_set_hash(skb, get_unaligned_be32(hash), PKT_HASH_TYPE_L4);
 		break;
 	}
 }
 
+static void
+nfp_net_set_hash_desc(struct net_device *netdev, struct sk_buff *skb,
+		      struct nfp_net_rx_desc *rxd)
+{
+	struct nfp_net_rx_hash *rx_hash;
+
+	if (!(rxd->rxd.flags & PCIE_DESC_RX_RSS))
+		return;
+
+	rx_hash = (struct nfp_net_rx_hash *)(skb->data - sizeof(*rx_hash));
+
+	nfp_net_set_hash(netdev, skb, get_unaligned_be32(&rx_hash->hash_type),
+			 &rx_hash->hash);
+}
+
+static void *
+nfp_net_parse_meta(struct net_device *netdev, struct sk_buff *skb,
+		   int meta_len)
+{
+	u8 *data = skb->data - meta_len;
+	u32 meta_info;
+
+	meta_info = get_unaligned_be32(data);
+	data += 4;
+
+	while (meta_info) {
+		switch (meta_info & NFP_NET_META_FIELD_MASK) {
+		case NFP_NET_META_HASH:
+			meta_info >>= NFP_NET_META_FIELD_SIZE;
+			nfp_net_set_hash(netdev, skb,
+					 meta_info & NFP_NET_META_FIELD_MASK,
+					 (__be32 *)data);
+			data += 4;
+			break;
+		case NFP_NET_META_MARK:
+			skb->mark = get_unaligned_be32(data);
+			data += 4;
+			break;
+		default:
+			return NULL;
+		}
+
+		meta_info >>= NFP_NET_META_FIELD_SIZE;
+	}
+
+	return data;
+}
+
 /**
  * nfp_net_rx() - receive up to @budget packets on @rx_ring
  * @rx_ring:   RX ring to receive from
@@ -1439,14 +1473,29 @@ static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget)
 			skb_reserve(skb, nn->rx_offset);
 		skb_put(skb, data_len - meta_len);
 
-		nfp_net_set_hash(nn->netdev, skb, rxd);
-
 		/* Stats update */
 		u64_stats_update_begin(&r_vec->rx_sync);
 		r_vec->rx_pkts++;
 		r_vec->rx_bytes += skb->len;
 		u64_stats_update_end(&r_vec->rx_sync);
 
+		if (nn->fw_ver.major <= 3) {
+			nfp_net_set_hash_desc(nn->netdev, skb, rxd);
+		} else if (meta_len) {
+			void *end;
+
+			end = nfp_net_parse_meta(nn->netdev, skb, meta_len);
+			if (unlikely(end != skb->data)) {
+				u64_stats_update_begin(&r_vec->rx_sync);
+				r_vec->rx_drops++;
+				u64_stats_update_end(&r_vec->rx_sync);
+
+				dev_kfree_skb_any(skb);
+				nn_warn_ratelimit(nn, "invalid RX packet metadata\n");
+				continue;
+			}
+		}
+
 		skb_record_rx_queue(skb, rx_ring->idx);
 		skb->protocol = eth_type_trans(skb, nn->netdev);
 
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h b/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h
index 7aa11f3c5ef0..93b10b441acb 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h
@@ -66,6 +66,13 @@
 #define NFP_NET_LSO_MAX_HDR_SZ		255
 
 /**
+ * Prepend field types
+ */
+#define NFP_NET_META_FIELD_SIZE		4
+#define NFP_NET_META_HASH		1 /* next field carries hash type */
+#define NFP_NET_META_MARK		2
+
+/**
  * Hash type pre-pended when a RSS hash was computed
  */
 #define NFP_NET_RSS_NONE                0
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_netvf_main.c b/drivers/net/ethernet/netronome/nfp/nfp_netvf_main.c
index f7062cb648e1..2800bbf65a89 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_netvf_main.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_netvf_main.c
@@ -148,7 +148,7 @@ static int nfp_netvf_pci_probe(struct pci_dev *pdev,
 		dev_warn(&pdev->dev, "OBSOLETE Firmware detected - VF isolation not available\n");
 	} else {
 		switch (fw_ver.major) {
-		case 1 ... 3:
+		case 1 ... 4:
 			if (is_nfp3200) {
 				stride = 2;
 				tx_bar_no = NFP_NET_Q0_BAR;
-- 
1.9.1

[PATCHv5 net-next 13/15] net: act_mirred: allow statistic updates from offloaded actions

[Jakub Kicinski]

Implement .stats_update() callback.  The implementation
is generic and can be reused by other simple actions if
needed.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 net/sched/act_mirred.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/net/sched/act_mirred.c b/net/sched/act_mirred.c
index 6038c85d92f5..f9862d89cb93 100644
--- a/net/sched/act_mirred.c
+++ b/net/sched/act_mirred.c
@@ -204,6 +204,13 @@ static int tcf_mirred(struct sk_buff *skb, const struct tc_action *a,
 	return retval;
 }
 
+static void tcf_stats_update(struct tc_action *a, u64 bytes, u32 packets,
+			     u64 lastuse)
+{
+	tcf_lastuse_update(&a->tcfa_tm);
+	_bstats_cpu_update(this_cpu_ptr(a->cpu_bstats), bytes, packets);
+}
+
 static int tcf_mirred_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 {
 	unsigned char *b = skb_tail_pointer(skb);
@@ -280,6 +287,7 @@ static struct tc_action_ops act_mirred_ops = {
 	.type		=	TCA_ACT_MIRRED,
 	.owner		=	THIS_MODULE,
 	.act		=	tcf_mirred,
+	.stats_update	=	tcf_stats_update,
 	.dump		=	tcf_mirred_dump,
 	.cleanup	=	tcf_mirred_release,
 	.init		=	tcf_mirred_init,
-- 
1.9.1

[PATCHv5 net-next 14/15] nfp: bpf: add support for legacy redirect action

[Jakub Kicinski]

Data path has redirect support so expressing redirect
to the port frame came from is a trivial matter of
setting the right result code.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_bpf.h         | 1 +
 drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c     | 2 ++
 drivers/net/ethernet/netronome/nfp/nfp_net_offload.c | 4 ++++
 3 files changed, 7 insertions(+)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_bpf.h b/drivers/net/ethernet/netronome/nfp/nfp_bpf.h
index d550fbc4768a..378d3c35cad5 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_bpf.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_bpf.h
@@ -60,6 +60,7 @@ enum static_regs {
 
 enum nfp_bpf_action_type {
 	NN_ACT_TC_DROP,
+	NN_ACT_TC_REDIR,
 };
 
 /* Software register representation, hardware encoding in asm.h */
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c b/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
index 42a8afb67fc8..60a99e0bf459 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
@@ -1440,6 +1440,7 @@ static void nfp_outro_tc_legacy(struct nfp_prog *nfp_prog)
 {
 	const u8 act2code[] = {
 		[NN_ACT_TC_DROP]  = 0x22,
+		[NN_ACT_TC_REDIR] = 0x24
 	};
 	/* Target for aborts */
 	nfp_prog->tgt_abort = nfp_prog_current_offset(nfp_prog);
@@ -1468,6 +1469,7 @@ static void nfp_outro(struct nfp_prog *nfp_prog)
 {
 	switch (nfp_prog->act) {
 	case NN_ACT_TC_DROP:
+	case NN_ACT_TC_REDIR:
 		nfp_outro_tc_legacy(nfp_prog);
 		break;
 	}
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c b/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
index 0537a53e2174..1ec8e5b74651 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
@@ -123,6 +123,10 @@ nfp_net_bpf_get_act(struct nfp_net *nn, struct tc_cls_bpf_offload *cls_bpf)
 	list_for_each_entry(a, &actions, list) {
 		if (is_tcf_gact_shot(a))
 			return NN_ACT_TC_DROP;
+
+		if (is_tcf_mirred_redirect(a) &&
+		    tcf_mirred_ifindex(a) == nn->netdev->ifindex)
+			return NN_ACT_TC_REDIR;
 	}
 
 	return -ENOTSUPP;
-- 
1.9.1

[PATCHv5 net-next 15/15] nfp: bpf: add offload of TC direct action mode

[Jakub Kicinski]

Add offload of TC in direct action mode.  We just need
to provide appropriate checks in the verifier and
a new outro block to translate the exit codes to what
data path expects

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_bpf.h       |  1 +
 drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c   | 66 ++++++++++++++++++++++
 .../net/ethernet/netronome/nfp/nfp_bpf_verifier.c  | 11 +++-
 .../net/ethernet/netronome/nfp/nfp_net_offload.c   |  6 +-
 4 files changed, 82 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_bpf.h b/drivers/net/ethernet/netronome/nfp/nfp_bpf.h
index 378d3c35cad5..25e9bd885f6f 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_bpf.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_bpf.h
@@ -61,6 +61,7 @@ enum static_regs {
 enum nfp_bpf_action_type {
 	NN_ACT_TC_DROP,
 	NN_ACT_TC_REDIR,
+	NN_ACT_DIRECT,
 };
 
 /* Software register representation, hardware encoding in asm.h */
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c b/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
index 60a99e0bf459..ca741763b7e9 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
@@ -321,6 +321,16 @@ __emit_br(struct nfp_prog *nfp_prog, enum br_mask mask, enum br_ev_pip ev_pip,
 	nfp_prog_push(nfp_prog, insn);
 }
 
+static void emit_br_def(struct nfp_prog *nfp_prog, u16 addr, u8 defer)
+{
+	if (defer > 2) {
+		pr_err("BUG: branch defer out of bounds %d\n", defer);
+		nfp_prog->error = -EFAULT;
+		return;
+	}
+	__emit_br(nfp_prog, BR_UNC, BR_EV_PIP_UNCOND, BR_CSS_NONE, addr, defer);
+}
+
 static void
 emit_br(struct nfp_prog *nfp_prog, enum br_mask mask, u16 addr, u8 defer)
 {
@@ -1465,9 +1475,65 @@ static void nfp_outro_tc_legacy(struct nfp_prog *nfp_prog)
 		      SHF_SC_L_SHF, 16);
 }
 
+static void nfp_outro_tc_da(struct nfp_prog *nfp_prog)
+{
+	/* TC direct-action mode:
+	 *   0,1   ok        NOT SUPPORTED[1]
+	 *   2   drop  0x22 -> drop,  count as stat1
+	 *   4,5 nuke  0x02 -> drop
+	 *   7  redir  0x44 -> redir, count as stat2
+	 *   * unspec  0x11 -> pass,  count as stat0
+	 *
+	 * [1] We can't support OK and RECLASSIFY because we can't tell TC
+	 *     the exact decision made.  We are forced to support UNSPEC
+	 *     to handle aborts so that's the only one we handle for passing
+	 *     packets up the stack.
+	 */
+	/* Target for aborts */
+	nfp_prog->tgt_abort = nfp_prog_current_offset(nfp_prog);
+
+	emit_br_def(nfp_prog, nfp_prog->tgt_done, 2);
+
+	emit_alu(nfp_prog, reg_a(0),
+		 reg_none(), ALU_OP_NONE, NFP_BPF_ABI_FLAGS);
+	emit_ld_field(nfp_prog, reg_a(0), 0xc, reg_imm(0x11), SHF_SC_L_SHF, 16);
+
+	/* Target for normal exits */
+	nfp_prog->tgt_out = nfp_prog_current_offset(nfp_prog);
+
+	/* if R0 > 7 jump to abort */
+	emit_alu(nfp_prog, reg_none(), reg_imm(7), ALU_OP_SUB, reg_b(0));
+	emit_br(nfp_prog, BR_BLO, nfp_prog->tgt_abort, 0);
+	emit_alu(nfp_prog, reg_a(0),
+		 reg_none(), ALU_OP_NONE, NFP_BPF_ABI_FLAGS);
+
+	wrp_immed(nfp_prog, reg_b(2), 0x41221211);
+	wrp_immed(nfp_prog, reg_b(3), 0x41001211);
+
+	emit_shf(nfp_prog, reg_a(1),
+		 reg_none(), SHF_OP_NONE, reg_b(0), SHF_SC_L_SHF, 2);
+
+	emit_alu(nfp_prog, reg_none(), reg_a(1), ALU_OP_OR, reg_imm(0));
+	emit_shf(nfp_prog, reg_a(2),
+		 reg_imm(0xf), SHF_OP_AND, reg_b(2), SHF_SC_R_SHF, 0);
+
+	emit_alu(nfp_prog, reg_none(), reg_a(1), ALU_OP_OR, reg_imm(0));
+	emit_shf(nfp_prog, reg_b(2),
+		 reg_imm(0xf), SHF_OP_AND, reg_b(3), SHF_SC_R_SHF, 0);
+
+	emit_br_def(nfp_prog, nfp_prog->tgt_done, 2);
+
+	emit_shf(nfp_prog, reg_b(2),
+		 reg_a(2), SHF_OP_OR, reg_b(2), SHF_SC_L_SHF, 4);
+	emit_ld_field(nfp_prog, reg_a(0), 0xc, reg_b(2), SHF_SC_L_SHF, 16);
+}
+
 static void nfp_outro(struct nfp_prog *nfp_prog)
 {
 	switch (nfp_prog->act) {
+	case NN_ACT_DIRECT:
+		nfp_outro_tc_da(nfp_prog);
+		break;
 	case NN_ACT_TC_DROP:
 	case NN_ACT_TC_REDIR:
 		nfp_outro_tc_legacy(nfp_prog);
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_bpf_verifier.c b/drivers/net/ethernet/netronome/nfp/nfp_bpf_verifier.c
index 15c460964810..af2d7c8bd8bf 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_bpf_verifier.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_bpf_verifier.c
@@ -86,7 +86,16 @@ nfp_bpf_check_exit(struct nfp_prog *nfp_prog,
 		return -EINVAL;
 	}
 
-	if (reg0->imm != 0 && (reg0->imm & ~0U) != ~0U) {
+	if (nfp_prog->act != NN_ACT_DIRECT &&
+	    reg0->imm != 0 && (reg0->imm & ~0U) != ~0U) {
+		pr_info("unsupported exit state: %d, imm: %llx\n",
+			reg0->type, reg0->imm);
+		return -EINVAL;
+	}
+
+	if (nfp_prog->act == NN_ACT_DIRECT && reg0->imm <= TC_ACT_REDIRECT &&
+	    reg0->imm != TC_ACT_SHOT && reg0->imm != TC_ACT_STOLEN &&
+	    reg0->imm != TC_ACT_QUEUED) {
 		pr_info("unsupported exit state: %d, imm: %llx\n",
 			reg0->type, reg0->imm);
 		return -EINVAL;
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c b/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
index 1ec8e5b74651..43f42f842eda 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
@@ -112,8 +112,12 @@ nfp_net_bpf_get_act(struct nfp_net *nn, struct tc_cls_bpf_offload *cls_bpf)
 	LIST_HEAD(actions);
 
 	/* TC direct action */
-	if (cls_bpf->exts_integrated)
+	if (cls_bpf->exts_integrated) {
+		if (tc_no_actions(cls_bpf->exts))
+			return NN_ACT_DIRECT;
+
 		return -ENOTSUPP;
+	}
 
 	/* TC legacy mode */
 	if (!tc_single_action(cls_bpf->exts))
-- 
1.9.1