nfp: bpf: support setting the RX queue index

BPF has access to all internal FW datapath structures.  Including
the structure containing RX queue selection.  With little coordination
with the datapath we can let the offloaded BPF select the RX queue.
We just need a way to tell the datapath that queue selection has already
been done and it shouldn't overwrite it.  Define a bit to tell datapath
BPF already selected a queue (QSEL_SET), if the selected queue is not
enabled (>= number of enabled queues) datapath will perform normal RSS.

BPF queue selection on the NIC can be used to replace standard
datapath RSS with fully programmable BPF/XDP RSS.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

drivers/net/ethernet/netronome/nfp/bpf/fw.h

@@ -50,6 +50,7 @@ enum bpf_cap_tlv_type {
 	NFP_BPF_CAP_TYPE_ADJUST_HEAD	= 2,
 	NFP_BPF_CAP_TYPE_MAPS		= 3,
 	NFP_BPF_CAP_TYPE_RANDOM		= 4,
+	NFP_BPF_CAP_TYPE_QUEUE_SELECT	= 5,
 };
 
 struct nfp_bpf_cap_tlv_func {

drivers/net/ethernet/netronome/nfp/bpf/jit.c

@@ -42,6 +42,7 @@
 
 #include "main.h"
 #include "../nfp_asm.h"
+#include "../nfp_net_ctrl.h"
 
 /* --- NFP prog --- */
 /* Foreach "multiple" entries macros provide pos and next<n> pointers.
@@ -1470,6 +1471,38 @@ nfp_perf_event_output(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 	return 0;
 }
 
+static int
+nfp_queue_select(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	u32 jmp_tgt;
+
+	jmp_tgt = nfp_prog_current_offset(nfp_prog) + 5;
+
+	/* Make sure the queue id fits into FW field */
+	emit_alu(nfp_prog, reg_none(), reg_a(meta->insn.src_reg * 2),
+		 ALU_OP_AND_NOT_B, reg_imm(0xff));
+	emit_br(nfp_prog, BR_BEQ, jmp_tgt, 2);
+
+	/* Set the 'queue selected' bit and the queue value */
+	emit_shf(nfp_prog, pv_qsel_set(nfp_prog),
+		 pv_qsel_set(nfp_prog), SHF_OP_OR, reg_imm(1),
+		 SHF_SC_L_SHF, PKT_VEL_QSEL_SET_BIT);
+	emit_ld_field(nfp_prog,
+		      pv_qsel_val(nfp_prog), 0x1, reg_b(meta->insn.src_reg * 2),
+		      SHF_SC_NONE, 0);
+	/* Delay slots end here, we will jump over next instruction if queue
+	 * value fits into the field.
+	 */
+	emit_ld_field(nfp_prog,
+		      pv_qsel_val(nfp_prog), 0x1, reg_imm(NFP_NET_RXR_MAX),
+		      SHF_SC_NONE, 0);
+
+	if (!nfp_prog_confirm_current_offset(nfp_prog, jmp_tgt))
+		return -EINVAL;
+
+	return 0;
+}
+
 /* --- Callbacks --- */
 static int mov_reg64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 {
@@ -2160,6 +2193,17 @@ mem_stx_stack(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
 			    false, wrp_lmem_store);
 }
 
+static int mem_stx_xdp(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
+{
+	switch (meta->insn.off) {
+	case offsetof(struct xdp_md, rx_queue_index):
+		return nfp_queue_select(nfp_prog, meta);
+	}
+
+	WARN_ON_ONCE(1); /* verifier should have rejected bad accesses */
+	return -EOPNOTSUPP;
+}
+
 static int
 mem_stx(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
 	unsigned int size)
@@ -2186,6 +2230,9 @@ static int mem_stx2(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 
 static int mem_stx4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta)
 {
+	if (meta->ptr.type == PTR_TO_CTX)
+		if (nfp_prog->type == BPF_PROG_TYPE_XDP)
+			return mem_stx_xdp(nfp_prog, meta);
 	return mem_stx(nfp_prog, meta, 4);
 }
 

drivers/net/ethernet/netronome/nfp/bpf/main.c

@@ -334,6 +334,13 @@ nfp_bpf_parse_cap_random(struct nfp_app_bpf *bpf, void __iomem *value,
 	return 0;
 }
 
+static int
+nfp_bpf_parse_cap_qsel(struct nfp_app_bpf *bpf, void __iomem *value, u32 length)
+{
+	bpf->queue_select = true;
+	return 0;
+}
+
 static int nfp_bpf_parse_capabilities(struct nfp_app *app)
 {
 	struct nfp_cpp *cpp = app->pf->cpp;
@@ -376,6 +383,10 @@ static int nfp_bpf_parse_capabilities(struct nfp_app *app)
 			if (nfp_bpf_parse_cap_random(app->priv, value, length))
 				goto err_release_free;
 			break;
+		case NFP_BPF_CAP_TYPE_QUEUE_SELECT:
+			if (nfp_bpf_parse_cap_qsel(app->priv, value, length))
+				goto err_release_free;
+			break;
 		default:
 			nfp_dbg(cpp, "unknown BPF capability: %d\n", type);
 			break;

drivers/net/ethernet/netronome/nfp/bpf/main.h

@@ -82,10 +82,16 @@ enum static_regs {
 enum pkt_vec {
 	PKT_VEC_PKT_LEN		= 0,
 	PKT_VEC_PKT_PTR		= 2,
+	PKT_VEC_QSEL_SET	= 4,
+	PKT_VEC_QSEL_VAL	= 6,
 };
 
+#define PKT_VEL_QSEL_SET_BIT	4
+
 #define pv_len(np)	reg_lm(1, PKT_VEC_PKT_LEN)
 #define pv_ctm_ptr(np)	reg_lm(1, PKT_VEC_PKT_PTR)
+#define pv_qsel_set(np)	reg_lm(1, PKT_VEC_QSEL_SET)
+#define pv_qsel_val(np)	reg_lm(1, PKT_VEC_QSEL_VAL)
 
 #define stack_reg(np)	reg_a(STATIC_REG_STACK)
 #define stack_imm(np)	imm_b(np)
@@ -139,6 +145,7 @@ enum pkt_vec {
  * @helpers.perf_event_output:	output perf event to a ring buffer
  *
  * @pseudo_random:	FW initialized the pseudo-random machinery (CSRs)
+ * @queue_select:	BPF can set the RX queue ID in packet vector
  */
 struct nfp_app_bpf {
 	struct nfp_app *app;
@@ -181,6 +188,7 @@ struct nfp_app_bpf {
 	} helpers;
 
 	bool pseudo_random;
+	bool queue_select;
 };
 
 enum nfp_bpf_map_use {

drivers/net/ethernet/netronome/nfp/bpf/verifier.c

@@ -467,6 +467,30 @@ nfp_bpf_check_ptr(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
 	return 0;
 }
 
+static int
+nfp_bpf_check_store(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
+		    struct bpf_verifier_env *env)
+{
+	const struct bpf_reg_state *reg = cur_regs(env) + meta->insn.dst_reg;
+
+	if (reg->type == PTR_TO_CTX) {
+		if (nfp_prog->type == BPF_PROG_TYPE_XDP) {
+			/* XDP ctx accesses must be 4B in size */
+			switch (meta->insn.off) {
+			case offsetof(struct xdp_md, rx_queue_index):
+				if (nfp_prog->bpf->queue_select)
+					goto exit_check_ptr;
+				pr_vlog(env, "queue selection not supported by FW\n");
+				return -EOPNOTSUPP;
+			}
+		}
+		pr_vlog(env, "unsupported store to context field\n");
+		return -EOPNOTSUPP;
+	}
+exit_check_ptr:
+	return nfp_bpf_check_ptr(nfp_prog, meta, env, meta->insn.dst_reg);
+}
+
 static int
 nfp_bpf_check_xadd(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
 		   struct bpf_verifier_env *env)
@@ -522,8 +546,8 @@ nfp_verify_insn(struct bpf_verifier_env *env, int insn_idx, int prev_insn_idx)
 		return nfp_bpf_check_ptr(nfp_prog, meta, env,
 					 meta->insn.src_reg);
 	if (is_mbpf_store(meta))
-		return nfp_bpf_check_ptr(nfp_prog, meta, env,
-					 meta->insn.dst_reg);
+		return nfp_bpf_check_store(nfp_prog, meta, env);
+
 	if (is_mbpf_xadd(meta))
 		return nfp_bpf_check_xadd(nfp_prog, meta, env);
 

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

@@ -183,16 +183,18 @@ enum shf_sc {
 #define OP_ALU_DST_LMEXTN	0x80000000000ULL
 
 enum alu_op {
-	ALU_OP_NONE	= 0x00,
-	ALU_OP_ADD	= 0x01,
-	ALU_OP_NOT	= 0x04,
-	ALU_OP_ADD_2B	= 0x05,
-	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,
+	ALU_OP_NONE		= 0x00,
+	ALU_OP_ADD		= 0x01,
+	ALU_OP_NOT		= 0x04,
+	ALU_OP_ADD_2B		= 0x05,
+	ALU_OP_AND		= 0x08,
+	ALU_OP_AND_NOT_A	= 0x0c,
+	ALU_OP_SUB_C		= 0x0d,
+	ALU_OP_AND_NOT_B	= 0x10,
+	ALU_OP_ADD_C		= 0x11,
+	ALU_OP_OR		= 0x14,
+	ALU_OP_SUB		= 0x15,
+	ALU_OP_XOR		= 0x18,
 };
 
 enum alu_dst_ab {