Merge branch 'bpf-event-output-offload'

Jakub Kicinski says:

====================
This series centres on NFP offload of bpf_event_output().  The
first patch allows perf event arrays to be used by offloaded
programs.  Next patch makes the nfp driver keep track of such
arrays to be able to filter FW events referring to maps.
Perf event arrays are not device bound.  Having driver
reimplement and manage the perf array seems brittle and unnecessary.

Patch 4 moves slightly the verifier step which replaces map fds
with map pointers.  This is useful for nfp JIT since we can then
easily replace host pointers with NFP table ids (patch 6).  This
allows us to lift the limitation on map helpers having to be used
with the same map pointer on all paths.  Second use of replacing
fds with real host map pointers is that we can use the host map
pointer as a key for FW events in perf event array offload.

Patch 5 adds perf event output offload support for the NFP.

There are some differences between bpf_event_output() offloaded
and non-offloaded version.  The FW messages which carry events
may get dropped and reordered relatively easily.  The return codes
from the helper are also not guaranteed to match the host.  Users
are warned about some of those discrepancies with a one time
warning message to kernel logs.

bpftool gains an ability to dump perf ring events in a very simple
format.  This was very useful for testing and simple debug, maybe
it will be useful to others?

Last patch is a trivial comment fix.
====================

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

8 files changed, 385 insertions, 31 deletions

diff --git a/drivers/net/ethernet/netronome/nfp/bpf/cmsg.c b/drivers/net/ethernet/netronome/nfp/bpf/cmsg.c
index 7e298148ca26..cb87fccb9f6a 100644
--- a/drivers/net/ethernet/netronome/nfp/bpf/cmsg.c
+++ b/drivers/net/ethernet/netronome/nfp/bpf/cmsg.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2017 Netronome Systems, Inc.
+ * Copyright (C) 2017-2018 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
@@ -102,6 +102,15 @@ nfp_bpf_cmsg_map_req_alloc(struct nfp_app_bpf *bpf, unsigned int n)
 	return nfp_bpf_cmsg_alloc(bpf, size);
 }
 
+static u8 nfp_bpf_cmsg_get_type(struct sk_buff *skb)
+{
+	struct cmsg_hdr *hdr;
+
+	hdr = (struct cmsg_hdr *)skb->data;
+
+	return hdr->type;
+}
+
 static unsigned int nfp_bpf_cmsg_get_tag(struct sk_buff *skb)
 {
 	struct cmsg_hdr *hdr;
@@ -431,6 +440,11 @@ void nfp_bpf_ctrl_msg_rx(struct nfp_app *app, struct sk_buff *skb)
 		goto err_free;
 	}
 
+	if (nfp_bpf_cmsg_get_type(skb) == CMSG_TYPE_BPF_EVENT) {
+		nfp_bpf_event_output(bpf, skb);
+		return;
+	}
+
 	nfp_ctrl_lock(bpf->app->ctrl);
 
 	tag = nfp_bpf_cmsg_get_tag(skb);
diff --git a/drivers/net/ethernet/netronome/nfp/bpf/fw.h b/drivers/net/ethernet/netronome/nfp/bpf/fw.h
index 39639ac28b01..3dbc21653ce5 100644
--- a/drivers/net/ethernet/netronome/nfp/bpf/fw.h
+++ b/drivers/net/ethernet/netronome/nfp/bpf/fw.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2017 Netronome Systems, Inc.
+ * Copyright (C) 2017-2018 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
@@ -37,6 +37,14 @@
 #include <linux/bitops.h>
 #include <linux/types.h>
 
+/* Kernel's enum bpf_reg_type is not uABI so people may change it breaking
+ * our FW ABI.  In that case we will do translation in the driver.
+ */
+#define NFP_BPF_SCALAR_VALUE		1
+#define NFP_BPF_MAP_VALUE		4
+#define NFP_BPF_STACK			6
+#define NFP_BPF_PACKET_DATA		8
+
 enum bpf_cap_tlv_type {
 	NFP_BPF_CAP_TYPE_FUNC		= 1,
 	NFP_BPF_CAP_TYPE_ADJUST_HEAD	= 2,
@@ -81,6 +89,7 @@ enum nfp_bpf_cmsg_type {
 	CMSG_TYPE_MAP_DELETE	= 5,
 	CMSG_TYPE_MAP_GETNEXT	= 6,
 	CMSG_TYPE_MAP_GETFIRST	= 7,
+	CMSG_TYPE_BPF_EVENT	= 8,
 	__CMSG_TYPE_MAP_MAX,
 };
 
@@ -155,4 +164,13 @@ struct cmsg_reply_map_op {
 	__be32 resv;
 	struct cmsg_key_value_pair elem[0];
 };
+
+struct cmsg_bpf_event {
+	struct cmsg_hdr hdr;
+	__be32 cpu_id;
+	__be64 map_ptr;
+	__be32 data_size;
+	__be32 pkt_size;
+	u8 data[0];
+};
 #endif
diff --git a/drivers/net/ethernet/netronome/nfp/bpf/jit.c b/drivers/net/ethernet/netronome/nfp/bpf/jit.c
index 65f0791cae0c..326a2085d650 100644
--- a/drivers/net/ethernet/netronome/nfp/bpf/jit.c
+++ b/drivers/net/ethernet/netronome/nfp/bpf/jit.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2016-2017 Netronome Systems, Inc.
+ * Copyright (C) 2016-2018 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
@@ -1395,15 +1395,9 @@ static int adjust_head(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 static int
 map_call_stack_common(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 {
-	struct bpf_offloaded_map *offmap;
-	struct nfp_bpf_map *nfp_map;
 	bool load_lm_ptr;
 	u32 ret_tgt;
 	s64 lm_off;
-	swreg tid;
-
-	offmap = (struct bpf_offloaded_map *)meta->arg1.map_ptr;
-	nfp_map = offmap->dev_priv;
 
 	/* We only have to reload LM0 if the key is not at start of stack */
 	lm_off = nfp_prog->stack_depth;
@@ -1416,17 +1410,12 @@ map_call_stack_common(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 	if (meta->func_id == BPF_FUNC_map_update_elem)
 		emit_csr_wr(nfp_prog, reg_b(3 * 2), NFP_CSR_ACT_LM_ADDR2);
 
-	/* Load map ID into a register, it should actually fit as an immediate
-	 * but in case it doesn't deal with it here, not in the delay slots.
-	 */
-	tid = ur_load_imm_any(nfp_prog, nfp_map->tid, imm_a(nfp_prog));
-
 	emit_br_relo(nfp_prog, BR_UNC, BR_OFF_RELO + meta->func_id,
 		     2, RELO_BR_HELPER);
 	ret_tgt = nfp_prog_current_offset(nfp_prog) + 2;
 
 	/* Load map ID into A0 */
-	wrp_mov(nfp_prog, reg_a(0), tid);
+	wrp_mov(nfp_prog, reg_a(0), reg_a(2));
 
 	/* Load the return address into B0 */
 	wrp_immed_relo(nfp_prog, reg_b(0), ret_tgt, RELO_IMMED_REL);
@@ -1456,6 +1445,31 @@ nfp_get_prandom_u32(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 	return 0;
 }
 
+static int
+nfp_perf_event_output(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	swreg ptr_type;
+	u32 ret_tgt;
+
+	ptr_type = ur_load_imm_any(nfp_prog, meta->arg1.type, imm_a(nfp_prog));
+
+	ret_tgt = nfp_prog_current_offset(nfp_prog) + 3;
+
+	emit_br_relo(nfp_prog, BR_UNC, BR_OFF_RELO + meta->func_id,
+		     2, RELO_BR_HELPER);
+
+	/* Load ptr type into A1 */
+	wrp_mov(nfp_prog, reg_a(1), ptr_type);
+
+	/* Load the return address into B0 */
+	wrp_immed_relo(nfp_prog, reg_b(0), ret_tgt, RELO_IMMED_REL);
+
+	if (!nfp_prog_confirm_current_offset(nfp_prog, ret_tgt))
+		return -EINVAL;
+
+	return 0;
+}
+
 /* --- Callbacks --- */
 static int mov_reg64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 {
@@ -2411,6 +2425,8 @@ static int call(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 		return map_call_stack_common(nfp_prog, meta);
 	case BPF_FUNC_get_prandom_u32:
 		return nfp_get_prandom_u32(nfp_prog, meta);
+	case BPF_FUNC_perf_event_output:
+		return nfp_perf_event_output(nfp_prog, meta);
 	default:
 		WARN_ONCE(1, "verifier allowed unsupported function\n");
 		return -EOPNOTSUPP;
@@ -3227,6 +3243,33 @@ static int nfp_bpf_optimize(struct nfp_prog *nfp_prog)
 	return 0;
 }
 
+static int nfp_bpf_replace_map_ptrs(struct nfp_prog *nfp_prog)
+{
+	struct nfp_insn_meta *meta1, *meta2;
+	struct nfp_bpf_map *nfp_map;
+	struct bpf_map *map;
+
+	nfp_for_each_insn_walk2(nfp_prog, meta1, meta2) {
+		if (meta1->skip || meta2->skip)
+			continue;
+
+		if (meta1->insn.code != (BPF_LD | BPF_IMM | BPF_DW) ||
+		    meta1->insn.src_reg != BPF_PSEUDO_MAP_FD)
+			continue;
+
+		map = (void *)(unsigned long)((u32)meta1->insn.imm |
+					      (u64)meta2->insn.imm << 32);
+		if (bpf_map_offload_neutral(map))
+			continue;
+		nfp_map = map_to_offmap(map)->dev_priv;
+
+		meta1->insn.imm = nfp_map->tid;
+		meta2->insn.imm = 0;
+	}
+
+	return 0;
+}
+
 static int nfp_bpf_ustore_calc(u64 *prog, unsigned int len)
 {
 	__le64 *ustore = (__force __le64 *)prog;
@@ -3263,6 +3306,10 @@ int nfp_bpf_jit(struct nfp_prog *nfp_prog)
 {
 	int ret;
 
+	ret = nfp_bpf_replace_map_ptrs(nfp_prog);
+	if (ret)
+		return ret;
+
 	ret = nfp_bpf_optimize(nfp_prog);
 	if (ret)
 		return ret;
@@ -3353,6 +3400,9 @@ void *nfp_bpf_relo_for_vnic(struct nfp_prog *nfp_prog, struct nfp_bpf_vnic *bv)
 			case BPF_FUNC_map_delete_elem:
 				val = nfp_prog->bpf->helpers.map_delete;
 				break;
+			case BPF_FUNC_perf_event_output:
+				val = nfp_prog->bpf->helpers.perf_event_output;
+				break;
 			default:
 				pr_err("relocation of unknown helper %d\n",
 				       val);
diff --git a/drivers/net/ethernet/netronome/nfp/bpf/main.c b/drivers/net/ethernet/netronome/nfp/bpf/main.c
index 1dc424685f4e..d72f9e7f42da 100644
--- a/drivers/net/ethernet/netronome/nfp/bpf/main.c
+++ b/drivers/net/ethernet/netronome/nfp/bpf/main.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2017 Netronome Systems, Inc.
+ * Copyright (C) 2017-2018 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
@@ -43,6 +43,14 @@
 #include "fw.h"
 #include "main.h"
 
+const struct rhashtable_params nfp_bpf_maps_neutral_params = {
+	.nelem_hint		= 4,
+	.key_len		= FIELD_SIZEOF(struct nfp_bpf_neutral_map, ptr),
+	.key_offset		= offsetof(struct nfp_bpf_neutral_map, ptr),
+	.head_offset		= offsetof(struct nfp_bpf_neutral_map, l),
+	.automatic_shrinking	= true,
+};
+
 static bool nfp_net_ebpf_capable(struct nfp_net *nn)
 {
 #ifdef __LITTLE_ENDIAN
@@ -290,6 +298,9 @@ nfp_bpf_parse_cap_func(struct nfp_app_bpf *bpf, void __iomem *value, u32 length)
 	case BPF_FUNC_map_delete_elem:
 		bpf->helpers.map_delete = readl(&cap->func_addr);
 		break;
+	case BPF_FUNC_perf_event_output:
+		bpf->helpers.perf_event_output = readl(&cap->func_addr);
+		break;
 	}
 
 	return 0;
@@ -401,17 +412,28 @@ static int nfp_bpf_init(struct nfp_app *app)
 	init_waitqueue_head(&bpf->cmsg_wq);
 	INIT_LIST_HEAD(&bpf->map_list);
 
-	err = nfp_bpf_parse_capabilities(app);
+	err = rhashtable_init(&bpf->maps_neutral, &nfp_bpf_maps_neutral_params);
 	if (err)
 		goto err_free_bpf;
 
+	err = nfp_bpf_parse_capabilities(app);
+	if (err)
+		goto err_free_neutral_maps;
+
 	return 0;
 
+err_free_neutral_maps:
+	rhashtable_destroy(&bpf->maps_neutral);
 err_free_bpf:
 	kfree(bpf);
 	return err;
 }
 
+static void nfp_check_rhashtable_empty(void *ptr, void *arg)
+{
+	WARN_ON_ONCE(1);
+}
+
 static void nfp_bpf_clean(struct nfp_app *app)
 {
 	struct nfp_app_bpf *bpf = app->priv;
@@ -419,6 +441,8 @@ static void nfp_bpf_clean(struct nfp_app *app)
 	WARN_ON(!skb_queue_empty(&bpf->cmsg_replies));
 	WARN_ON(!list_empty(&bpf->map_list));
 	WARN_ON(bpf->maps_in_use || bpf->map_elems_in_use);
+	rhashtable_free_and_destroy(&bpf->maps_neutral,
+				    nfp_check_rhashtable_empty, NULL);
 	kfree(bpf);
 }
 
diff --git a/drivers/net/ethernet/netronome/nfp/bpf/main.h b/drivers/net/ethernet/netronome/nfp/bpf/main.h
index 68b5d326483d..82682378d57f 100644
--- a/drivers/net/ethernet/netronome/nfp/bpf/main.h
+++ b/drivers/net/ethernet/netronome/nfp/bpf/main.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2016-2017 Netronome Systems, Inc.
+ * Copyright (C) 2016-2018 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
@@ -39,6 +39,7 @@
 #include <linux/bpf_verifier.h>
 #include <linux/kernel.h>
 #include <linux/list.h>
+#include <linux/rhashtable.h>
 #include <linux/skbuff.h>
 #include <linux/types.h>
 #include <linux/wait.h>
@@ -114,6 +115,8 @@ enum pkt_vec {
  * @maps_in_use:	number of currently offloaded maps
  * @map_elems_in_use:	number of elements allocated to offloaded maps
  *
+ * @maps_neutral:	hash table of offload-neutral maps (on pointer)
+ *
  * @adjust_head:	adjust head capability
  * @adjust_head.flags:		extra flags for adjust head
  * @adjust_head.off_min:	minimal packet offset within buffer required
@@ -133,6 +136,7 @@ enum pkt_vec {
  * @helpers.map_lookup:		map lookup helper address
  * @helpers.map_update:		map update helper address
  * @helpers.map_delete:		map delete helper address
+ * @helpers.perf_event_output:	output perf event to a ring buffer
  *
  * @pseudo_random:	FW initialized the pseudo-random machinery (CSRs)
  */
@@ -150,6 +154,8 @@ struct nfp_app_bpf {
 	unsigned int maps_in_use;
 	unsigned int map_elems_in_use;
 
+	struct rhashtable maps_neutral;
+
 	struct nfp_bpf_cap_adjust_head {
 		u32 flags;
 		int off_min;
@@ -171,6 +177,7 @@ struct nfp_app_bpf {
 		u32 map_lookup;
 		u32 map_update;
 		u32 map_delete;
+		u32 perf_event_output;
 	} helpers;
 
 	bool pseudo_random;
@@ -199,6 +206,14 @@ struct nfp_bpf_map {
 	enum nfp_bpf_map_use use_map[];
 };
 
+struct nfp_bpf_neutral_map {
+	struct rhash_head l;
+	struct bpf_map *ptr;
+	u32 count;
+};
+
+extern const struct rhashtable_params nfp_bpf_maps_neutral_params;
+
 struct nfp_prog;
 struct nfp_insn_meta;
 typedef int (*instr_cb_t)(struct nfp_prog *, struct nfp_insn_meta *);
@@ -367,6 +382,8 @@ static inline bool is_mbpf_xadd(const struct nfp_insn_meta *meta)
  * @error: error code if something went wrong
  * @stack_depth: max stack depth from the verifier
  * @adjust_head_location: if program has single adjust head call - the insn no.
+ * @map_records_cnt: the number of map pointers recorded for this prog
+ * @map_records: the map record pointers from bpf->maps_neutral
  * @insns: list of BPF instruction wrappers (struct nfp_insn_meta)
  */
 struct nfp_prog {
@@ -390,6 +407,9 @@ struct nfp_prog {
 	unsigned int stack_depth;
 	unsigned int adjust_head_location;
 
+	unsigned int map_records_cnt;
+	struct nfp_bpf_neutral_map **map_records;
+
 	struct list_head insns;
 };
 
@@ -440,5 +460,7 @@ int nfp_bpf_ctrl_lookup_entry(struct bpf_offloaded_map *offmap,
 int nfp_bpf_ctrl_getnext_entry(struct bpf_offloaded_map *offmap,
 			       void *key, void *next_key);
 
+int nfp_bpf_event_output(struct nfp_app_bpf *bpf, struct sk_buff *skb);
+
 void nfp_bpf_ctrl_msg_rx(struct nfp_app *app, struct sk_buff *skb);
 #endif
diff --git a/drivers/net/ethernet/netronome/nfp/bpf/offload.c b/drivers/net/ethernet/netronome/nfp/bpf/offload.c
index 42d98792bd25..4db0ac1e42a8 100644
--- a/drivers/net/ethernet/netronome/nfp/bpf/offload.c
+++ b/drivers/net/ethernet/netronome/nfp/bpf/offload.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2016-2017 Netronome Systems, Inc.
+ * Copyright (C) 2016-2018 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
@@ -57,6 +57,126 @@
 #include "../nfp_net.h"
 
 static int
+nfp_map_ptr_record(struct nfp_app_bpf *bpf, struct nfp_prog *nfp_prog,
+		   struct bpf_map *map)
+{
+	struct nfp_bpf_neutral_map *record;
+	int err;
+
+	/* Map record paths are entered via ndo, update side is protected. */
+	ASSERT_RTNL();
+
+	/* Reuse path - other offloaded program is already tracking this map. */
+	record = rhashtable_lookup_fast(&bpf->maps_neutral, &map,
+					nfp_bpf_maps_neutral_params);
+	if (record) {
+		nfp_prog->map_records[nfp_prog->map_records_cnt++] = record;
+		record->count++;
+		return 0;
+	}
+
+	/* Grab a single ref to the map for our record.  The prog destroy ndo
+	 * happens after free_used_maps().
+	 */
+	map = bpf_map_inc(map, false);
+	if (IS_ERR(map))
+		return PTR_ERR(map);
+
+	record = kmalloc(sizeof(*record), GFP_KERNEL);
+	if (!record) {
+		err = -ENOMEM;
+		goto err_map_put;
+	}
+
+	record->ptr = map;
+	record->count = 1;
+
+	err = rhashtable_insert_fast(&bpf->maps_neutral, &record->l,
+				     nfp_bpf_maps_neutral_params);
+	if (err)
+		goto err_free_rec;
+
+	nfp_prog->map_records[nfp_prog->map_records_cnt++] = record;
+
+	return 0;
+
+err_free_rec:
+	kfree(record);
+err_map_put:
+	bpf_map_put(map);
+	return err;
+}
+
+static void
+nfp_map_ptrs_forget(struct nfp_app_bpf *bpf, struct nfp_prog *nfp_prog)
+{
+	bool freed = false;
+	int i;
+
+	ASSERT_RTNL();
+
+	for (i = 0; i < nfp_prog->map_records_cnt; i++) {
+		if (--nfp_prog->map_records[i]->count) {
+			nfp_prog->map_records[i] = NULL;
+			continue;
+		}
+
+		WARN_ON(rhashtable_remove_fast(&bpf->maps_neutral,
+					       &nfp_prog->map_records[i]->l,
+					       nfp_bpf_maps_neutral_params));
+		freed = true;
+	}
+
+	if (freed) {
+		synchronize_rcu();
+
+		for (i = 0; i < nfp_prog->map_records_cnt; i++)
+			if (nfp_prog->map_records[i]) {
+				bpf_map_put(nfp_prog->map_records[i]->ptr);
+				kfree(nfp_prog->map_records[i]);
+			}
+	}
+
+	kfree(nfp_prog->map_records);
+	nfp_prog->map_records = NULL;
+	nfp_prog->map_records_cnt = 0;
+}
+
+static int
+nfp_map_ptrs_record(struct nfp_app_bpf *bpf, struct nfp_prog *nfp_prog,
+		    struct bpf_prog *prog)
+{
+	int i, cnt, err;
+
+	/* Quickly count the maps we will have to remember */
+	cnt = 0;
+	for (i = 0; i < prog->aux->used_map_cnt; i++)
+		if (bpf_map_offload_neutral(prog->aux->used_maps[i]))
+			cnt++;
+	if (!cnt)
+		return 0;
+
+	nfp_prog->map_records = kmalloc_array(cnt,
+					      sizeof(nfp_prog->map_records[0]),
+					      GFP_KERNEL);
+	if (!nfp_prog->map_records)
+		return -ENOMEM;
+
+	for (i = 0; i < prog->aux->used_map_cnt; i++)
+		if (bpf_map_offload_neutral(prog->aux->used_maps[i])) {
+			err = nfp_map_ptr_record(bpf, nfp_prog,
+						 prog->aux->used_maps[i]);
+			if (err) {
+				nfp_map_ptrs_forget(bpf, nfp_prog);
+				return err;
+			}
+		}
+	WARN_ON(cnt != nfp_prog->map_records_cnt);
+
+	return 0;
+}
+
+static int
 nfp_prog_prepare(struct nfp_prog *nfp_prog, const struct bpf_insn *prog,
 		 unsigned int cnt)
 {
@@ -151,7 +271,7 @@ static int nfp_bpf_translate(struct nfp_net *nn, struct bpf_prog *prog)
 	prog->aux->offload->jited_len = nfp_prog->prog_len * sizeof(u64);
 	prog->aux->offload->jited_image = nfp_prog->prog;
 
-	return 0;
+	return nfp_map_ptrs_record(nfp_prog->bpf, nfp_prog, prog);
 }
 
 static int nfp_bpf_destroy(struct nfp_net *nn, struct bpf_prog *prog)
@@ -159,6 +279,7 @@ static int nfp_bpf_destroy(struct nfp_net *nn, struct bpf_prog *prog)
 	struct nfp_prog *nfp_prog = prog->aux->offload->dev_priv;
 
 	kvfree(nfp_prog->prog);
+	nfp_map_ptrs_forget(nfp_prog->bpf, nfp_prog);
 	nfp_prog_free(nfp_prog);
 
 	return 0;
@@ -320,6 +441,53 @@ int nfp_ndo_bpf(struct nfp_app *app, struct nfp_net *nn, struct netdev_bpf *bpf)
 	}
 }
 
+static unsigned long
+nfp_bpf_perf_event_copy(void *dst, const void *src,
+			unsigned long off, unsigned long len)
+{
+	memcpy(dst, src + off, len);
+	return 0;
+}
+
+int nfp_bpf_event_output(struct nfp_app_bpf *bpf, struct sk_buff *skb)
+{
+	struct cmsg_bpf_event *cbe = (void *)skb->data;
+	u32 pkt_size, data_size;
+	struct bpf_map *map;
+
+	if (skb->len < sizeof(struct cmsg_bpf_event))
+		goto err_drop;
+
+	pkt_size = be32_to_cpu(cbe->pkt_size);
+	data_size = be32_to_cpu(cbe->data_size);
+	map = (void *)(unsigned long)be64_to_cpu(cbe->map_ptr);
+
+	if (skb->len < sizeof(struct cmsg_bpf_event) + pkt_size + data_size)
+		goto err_drop;
+	if (cbe->hdr.ver != CMSG_MAP_ABI_VERSION)
+		goto err_drop;
+
+	rcu_read_lock();
+	if (!rhashtable_lookup_fast(&bpf->maps_neutral, &map,
+				    nfp_bpf_maps_neutral_params)) {
+		rcu_read_unlock();
+		pr_warn("perf event: dest map pointer %px not recognized, dropping event\n",
+			map);
+		goto err_drop;
+	}
+
+	bpf_event_output(map, be32_to_cpu(cbe->cpu_id),
+			 &cbe->data[round_up(pkt_size, 4)], data_size,
+			 cbe->data, pkt_size, nfp_bpf_perf_event_copy);
+	rcu_read_unlock();
+
+	dev_consume_skb_any(skb);
+	return 0;
+err_drop:
+	dev_kfree_skb_any(skb);
+	return -EINVAL;
+}
+
 static int
 nfp_net_bpf_load(struct nfp_net *nn, struct bpf_prog *prog,
 		 struct netlink_ext_ack *extack)
diff --git a/drivers/net/ethernet/netronome/nfp/bpf/verifier.c b/drivers/net/ethernet/netronome/nfp/bpf/verifier.c
index 06ad53ce4ad9..e163f3cfa47d 100644
--- a/drivers/net/ethernet/netronome/nfp/bpf/verifier.c
+++ b/drivers/net/ethernet/netronome/nfp/bpf/verifier.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2016-2017 Netronome Systems, Inc.
+ * Copyright (C) 2016-2018 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
@@ -36,6 +36,8 @@
 #include <linux/kernel.h>
 #include <linux/pkt_cls.h>
 
+#include "../nfp_app.h"
+#include "../nfp_main.h"
 #include "fw.h"
 #include "main.h"
 
@@ -149,15 +151,6 @@ nfp_bpf_map_call_ok(const char *fname, struct bpf_verifier_env *env,
 		return false;
 	}
 
-	/* Rest of the checks is only if we re-parse the same insn */
-	if (!meta->func_id)
-		return true;
-
-	if (meta->arg1.map_ptr != reg1->map_ptr) {
-		pr_vlog(env, "%s: called for different map\n", fname);
-		return false;
-	}
-
 	return true;
 }
 
@@ -216,6 +209,71 @@ nfp_bpf_check_call(struct nfp_prog *nfp_prog, struct bpf_verifier_env *env,
 		pr_vlog(env, "bpf_get_prandom_u32(): FW doesn't support random number generation\n");
 		return -EOPNOTSUPP;
 
+	case BPF_FUNC_perf_event_output:
+		BUILD_BUG_ON(NFP_BPF_SCALAR_VALUE != SCALAR_VALUE ||
+			     NFP_BPF_MAP_VALUE != PTR_TO_MAP_VALUE ||
+			     NFP_BPF_STACK != PTR_TO_STACK ||
+			     NFP_BPF_PACKET_DATA != PTR_TO_PACKET);
+
+		if (!bpf->helpers.perf_event_output) {
+			pr_vlog(env, "event_output: not supported by FW\n");
+			return -EOPNOTSUPP;
+		}
+
+		/* Force current CPU to make sure we can report the event
+		 * wherever we get the control message from FW.
+		 */
+		if (reg3->var_off.mask & BPF_F_INDEX_MASK ||
+		    (reg3->var_off.value & BPF_F_INDEX_MASK) !=
+		    BPF_F_CURRENT_CPU) {
+			char tn_buf[48];
+
+			tnum_strn(tn_buf, sizeof(tn_buf), reg3->var_off);
+			pr_vlog(env, "event_output: must use BPF_F_CURRENT_CPU, var_off: %s\n",
+				tn_buf);
+			return -EOPNOTSUPP;
+		}
+
+		/* Save space in meta, we don't care about arguments other
+		 * than 4th meta, shove it into arg1.
+		 */
+		reg1 = cur_regs(env) + BPF_REG_4;
+
+		if (reg1->type != SCALAR_VALUE /* NULL ptr */ &&
+		    reg1->type != PTR_TO_STACK &&
+		    reg1->type != PTR_TO_MAP_VALUE &&
+		    reg1->type != PTR_TO_PACKET) {
+			pr_vlog(env, "event_output: unsupported ptr type: %d\n",
+				reg1->type);
+			return -EOPNOTSUPP;
+		}
+
+		if (reg1->type == PTR_TO_STACK &&
+		    !nfp_bpf_stack_arg_ok("event_output", env, reg1, NULL))
+			return -EOPNOTSUPP;
+
+		/* Warn user that on offload NFP may return success even if map
+		 * is not going to accept the event, since the event output is
+		 * fully async and device won't know the state of the map.
+		 * There is also FW limitation on the event length.
+		 *
+		 * Lost events will not show up on the perf ring, driver
+		 * won't see them at all.  Events may also get reordered.
+		 */
+		dev_warn_once(&nfp_prog->bpf->app->pf->pdev->dev,
+			      "bpf: note: return codes and behavior of bpf_event_output() helper differs for offloaded programs!\n");
+		pr_vlog(env, "warning: return codes and behavior of event_output helper differ for offload!\n");
+
+		if (!meta->func_id)
+			break;
+
+		if (reg1->type != meta->arg1.type) {
+			pr_vlog(env, "event_output: ptr type changed: %d %d\n",
+				meta->arg1.type, reg1->type);
+			return -EINVAL;
+		}
+		break;
+
 	default:
 		pr_vlog(env, "unsupported function id: %d\n", func_id);
 		return -EOPNOTSUPP;
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_app.c b/drivers/net/ethernet/netronome/nfp/nfp_app.c
index 6aedef0ad433..0e0253c7e17b 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_app.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_app.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2017 Netronome Systems, Inc.
+ * Copyright (C) 2017-2018 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