diff options
Diffstat (limited to 'drivers/net/ethernet/netronome/nfp/bpf/jit.c')
| -rw-r--r-- | drivers/net/ethernet/netronome/nfp/bpf/jit.c | 2227 |
1 files changed, 1642 insertions, 585 deletions
diff --git a/drivers/net/ethernet/netronome/nfp/bpf/jit.c b/drivers/net/ethernet/netronome/nfp/bpf/jit.c index 239dfbe8a0a1..56451edf01c2 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 Netronome Systems, Inc. + * Copyright (C) 2016-2017 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 @@ -33,6 +33,7 @@ #define pr_fmt(fmt) "NFP net bpf: " fmt +#include <linux/bug.h> #include <linux/kernel.h> #include <linux/bpf.h> #include <linux/filter.h> @@ -66,28 +67,11 @@ 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) { @@ -101,162 +85,25 @@ static void nfp_prog_push(struct nfp_prog *nfp_prog, u64 insn) 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; - } + return nfp_prog->prog_len; } -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) +static bool +nfp_prog_confirm_current_offset(struct nfp_prog *nfp_prog, unsigned int off) { - 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, ®->i8); - reg->breg = nfp_swreg_to_rereg(lreg, false, has_imm8, ®->i8); - reg->swap = true; - } else { - reg->areg = nfp_swreg_to_rereg(lreg, false, has_imm8, ®->i8); - reg->breg = nfp_swreg_to_rereg(rreg, false, has_imm8, ®->i8); - } - - return 0; + /* If there is a recorded error we may have dropped instructions; + * that doesn't have to be due to translator bug, and the translation + * will fail anyway, so just return OK. + */ + if (nfp_prog->error) + return true; + return !WARN_ON_ONCE(nfp_prog_current_offset(nfp_prog) != off); } /* --- 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) + u8 mode, u8 xfer, u8 areg, u8 breg, u8 size, bool sync, bool indir) { enum cmd_ctx_swap ctx; u64 insn; @@ -274,14 +121,15 @@ __emit_cmd(struct nfp_prog *nfp_prog, enum cmd_tgt_map op, 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_INDIR, indir) | 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) +emit_cmd_any(struct nfp_prog *nfp_prog, enum cmd_tgt_map op, u8 mode, u8 xfer, + swreg lreg, swreg rreg, u8 size, bool sync, bool indir) { struct nfp_insn_re_regs reg; int err; @@ -296,8 +144,28 @@ emit_cmd(struct nfp_prog *nfp_prog, enum cmd_tgt_map op, nfp_prog->error = -EFAULT; return; } + if (reg.dst_lmextn || reg.src_lmextn) { + pr_err("cmd can't use LMextn\n"); + nfp_prog->error = -EFAULT; + return; + } + + __emit_cmd(nfp_prog, op, mode, xfer, reg.areg, reg.breg, size, sync, + indir); +} + +static void +emit_cmd(struct nfp_prog *nfp_prog, enum cmd_tgt_map op, u8 mode, u8 xfer, + swreg lreg, swreg rreg, u8 size, bool sync) +{ + emit_cmd_any(nfp_prog, op, mode, xfer, lreg, rreg, size, sync, false); +} - __emit_cmd(nfp_prog, op, mode, xfer, reg.areg, reg.breg, size, sync); +static void +emit_cmd_indir(struct nfp_prog *nfp_prog, enum cmd_tgt_map op, u8 mode, u8 xfer, + swreg lreg, swreg rreg, u8 size, bool sync) +{ + emit_cmd_any(nfp_prog, op, mode, xfer, lreg, rreg, size, sync, true); } static void @@ -321,68 +189,35 @@ __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) +static void +emit_br_relo(struct nfp_prog *nfp_prog, enum br_mask mask, u16 addr, u8 defer, + enum nfp_relo_type relo) { - if (defer > 2) { + if (mask == BR_UNC && 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) -{ __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); + nfp_prog->prog[nfp_prog->prog_len - 1] |= + FIELD_PREP(OP_RELO_TYPE, relo); } static void -emit_br_byte_neq(struct nfp_prog *nfp_prog, - u32 dst, u8 imm, u8 byte, u16 addr, u8 defer) +emit_br(struct nfp_prog *nfp_prog, enum br_mask mask, u16 addr, u8 defer) { - struct nfp_insn_re_regs reg; - int err; - - err = swreg_to_restricted(reg_none(), dst, reg_imm(imm), ®, true); - if (err) { - nfp_prog->error = err; - return; - } - - __emit_br_byte(nfp_prog, reg.areg, reg.breg, reg.i8, byte, false, addr, - defer); + emit_br_relo(nfp_prog, mask, addr, defer, RELO_BR_REL); } 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) + enum immed_shift shift, bool wr_both, + bool dst_lmextn, bool src_lmextn) { u64 insn; @@ -393,19 +228,21 @@ __emit_immed(struct nfp_prog *nfp_prog, u16 areg, u16 breg, u16 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); + FIELD_PREP(OP_IMMED_WR_AB, wr_both) | + FIELD_PREP(OP_IMMED_SRC_LMEXTN, src_lmextn) | + FIELD_PREP(OP_IMMED_DST_LMEXTN, dst_lmextn); nfp_prog_push(nfp_prog, insn); } static void -emit_immed(struct nfp_prog *nfp_prog, u32 dst, u16 imm, +emit_immed(struct nfp_prog *nfp_prog, swreg 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) { + if (swreg_type(dst) == NN_REG_IMM) { nfp_prog->error = -EFAULT; return; } @@ -416,14 +253,18 @@ emit_immed(struct nfp_prog *nfp_prog, u32 dst, u16 imm, return; } - __emit_immed(nfp_prog, reg.areg, reg.breg, imm >> 8, width, - invert, shift, reg.wr_both); + /* Use reg.dst when destination is No-Dest. */ + __emit_immed(nfp_prog, + swreg_type(dst) == NN_REG_NONE ? reg.dst : reg.areg, + reg.breg, imm >> 8, width, invert, shift, + reg.wr_both, reg.dst_lmextn, reg.src_lmextn); } 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) + u16 areg, enum shf_op op, u16 breg, bool i8, bool sw, bool wr_both, + bool dst_lmextn, bool src_lmextn) { u64 insn; @@ -445,14 +286,16 @@ __emit_shf(struct nfp_prog *nfp_prog, u16 dst, enum alu_dst_ab dst_ab, 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); + FIELD_PREP(OP_SHF_WR_AB, wr_both) | + FIELD_PREP(OP_SHF_SRC_LMEXTN, src_lmextn) | + FIELD_PREP(OP_SHF_DST_LMEXTN, dst_lmextn); 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) +emit_shf(struct nfp_prog *nfp_prog, swreg dst, + swreg lreg, enum shf_op op, swreg rreg, enum shf_sc sc, u8 shift) { struct nfp_insn_re_regs reg; int err; @@ -464,12 +307,14 @@ emit_shf(struct nfp_prog *nfp_prog, u32 dst, u32 lreg, enum shf_op op, u32 rreg, } __emit_shf(nfp_prog, reg.dst, reg.dst_ab, sc, shift, - reg.areg, op, reg.breg, reg.i8, reg.swap, reg.wr_both); + reg.areg, op, reg.breg, reg.i8, reg.swap, reg.wr_both, + reg.dst_lmextn, reg.src_lmextn); } 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) + u16 areg, enum alu_op op, u16 breg, bool swap, bool wr_both, + bool dst_lmextn, bool src_lmextn) { u64 insn; @@ -480,13 +325,16 @@ __emit_alu(struct nfp_prog *nfp_prog, u16 dst, enum alu_dst_ab dst_ab, 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); + FIELD_PREP(OP_ALU_WR_AB, wr_both) | + FIELD_PREP(OP_ALU_SRC_LMEXTN, src_lmextn) | + FIELD_PREP(OP_ALU_DST_LMEXTN, dst_lmextn); 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) +emit_alu(struct nfp_prog *nfp_prog, swreg dst, + swreg lreg, enum alu_op op, swreg rreg) { struct nfp_insn_ur_regs reg; int err; @@ -498,13 +346,15 @@ emit_alu(struct nfp_prog *nfp_prog, u32 dst, u32 lreg, enum alu_op op, u32 rreg) } __emit_alu(nfp_prog, reg.dst, reg.dst_ab, - reg.areg, op, reg.breg, reg.swap, reg.wr_both); + reg.areg, op, reg.breg, reg.swap, reg.wr_both, + reg.dst_lmextn, reg.src_lmextn); } 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) + bool zero, bool swap, bool wr_both, + bool dst_lmextn, bool src_lmextn) { u64 insn; @@ -517,33 +367,84 @@ __emit_ld_field(struct nfp_prog *nfp_prog, enum shf_sc sc, 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); + FIELD_PREP(OP_LDF_WR_AB, wr_both) | + FIELD_PREP(OP_LDF_SRC_LMEXTN, src_lmextn) | + FIELD_PREP(OP_LDF_DST_LMEXTN, dst_lmextn); 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) +emit_ld_field_any(struct nfp_prog *nfp_prog, swreg dst, u8 bmask, swreg src, + enum shf_sc sc, u8 shift, bool zero) { struct nfp_insn_re_regs reg; int err; - err = swreg_to_restricted(reg_none(), dst, src, ®, true); + /* Note: ld_field is special as it uses one of the src regs as dst */ + err = swreg_to_restricted(dst, dst, src, ®, 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); + reg.i8, zero, reg.swap, reg.wr_both, + reg.dst_lmextn, reg.src_lmextn); } static void -emit_ld_field(struct nfp_prog *nfp_prog, u32 dst, u8 bmask, u32 src, +emit_ld_field(struct nfp_prog *nfp_prog, swreg dst, u8 bmask, swreg src, enum shf_sc sc, u8 shift) { - emit_ld_field_any(nfp_prog, sc, shift, dst, bmask, src, false); + emit_ld_field_any(nfp_prog, dst, bmask, src, sc, shift, false); +} + +static void +__emit_lcsr(struct nfp_prog *nfp_prog, u16 areg, u16 breg, bool wr, u16 addr, + bool dst_lmextn, bool src_lmextn) +{ + u64 insn; + + insn = OP_LCSR_BASE | + FIELD_PREP(OP_LCSR_A_SRC, areg) | + FIELD_PREP(OP_LCSR_B_SRC, breg) | + FIELD_PREP(OP_LCSR_WRITE, wr) | + FIELD_PREP(OP_LCSR_ADDR, addr) | + FIELD_PREP(OP_LCSR_SRC_LMEXTN, src_lmextn) | + FIELD_PREP(OP_LCSR_DST_LMEXTN, dst_lmextn); + + nfp_prog_push(nfp_prog, insn); +} + +static void emit_csr_wr(struct nfp_prog *nfp_prog, swreg src, u16 addr) +{ + struct nfp_insn_ur_regs reg; + int err; + + /* This instruction takes immeds instead of reg_none() for the ignored + * operand, but we can't encode 2 immeds in one instr with our normal + * swreg infra so if param is an immed, we encode as reg_none() and + * copy the immed to both operands. + */ + if (swreg_type(src) == NN_REG_IMM) { + err = swreg_to_unrestricted(reg_none(), src, reg_none(), ®); + reg.breg = reg.areg; + } else { + err = swreg_to_unrestricted(reg_none(), src, reg_imm(0), ®); + } + if (err) { + nfp_prog->error = err; + return; + } + + __emit_lcsr(nfp_prog, reg.areg, reg.breg, true, addr / 4, + false, reg.src_lmextn); +} + +static void emit_nop(struct nfp_prog *nfp_prog) +{ + __emit_immed(nfp_prog, UR_REG_IMM, UR_REG_IMM, 0, 0, 0, 0, 0, 0, 0); } /* --- Wrappers --- */ @@ -565,7 +466,7 @@ static bool pack_immed(u32 imm, u16 *val, enum immed_shift *shift) return true; } -static void wrp_immed(struct nfp_prog *nfp_prog, u32 dst, u32 imm) +static void wrp_immed(struct nfp_prog *nfp_prog, swreg dst, u32 imm) { enum immed_shift shift; u16 val; @@ -582,11 +483,26 @@ static void wrp_immed(struct nfp_prog *nfp_prog, u32 dst, u32 imm) } } +static void +wrp_immed_relo(struct nfp_prog *nfp_prog, swreg dst, u32 imm, + enum nfp_relo_type relo) +{ + if (imm > 0xffff) { + pr_err("relocation of a large immediate!\n"); + nfp_prog->error = -EFAULT; + return; + } + emit_immed(nfp_prog, dst, imm, IMMED_WIDTH_ALL, false, IMMED_SHIFT_0B); + + nfp_prog->prog[nfp_prog->prog_len - 1] |= + FIELD_PREP(OP_RELO_TYPE, relo); +} + /* 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) +static swreg ur_load_imm_any(struct nfp_prog *nfp_prog, u32 imm, swreg tmp_reg) { if (FIELD_FIT(UR_REG_IMM_MAX, imm)) return reg_imm(imm); @@ -599,7 +515,7 @@ static u32 ur_load_imm_any(struct nfp_prog *nfp_prog, u32 imm, u32 tmp_reg) * 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) +static swreg re_load_imm_any(struct nfp_prog *nfp_prog, u32 imm, swreg tmp_reg) { if (FIELD_FIT(RE_REG_IMM_MAX, imm)) return reg_imm(imm); @@ -608,88 +524,565 @@ static u32 re_load_imm_any(struct nfp_prog *nfp_prog, u32 imm, u32 tmp_reg) return tmp_reg; } -static void -wrp_br_special(struct nfp_prog *nfp_prog, enum br_mask mask, - enum br_special special) +static void wrp_nops(struct nfp_prog *nfp_prog, unsigned int count) { - emit_br(nfp_prog, mask, 0, 0); + while (count--) + emit_nop(nfp_prog); +} - nfp_prog->prog[nfp_prog->prog_len - 1] |= - FIELD_PREP(OP_BR_SPECIAL, special); +static void wrp_mov(struct nfp_prog *nfp_prog, swreg dst, swreg src) +{ + emit_alu(nfp_prog, dst, reg_none(), ALU_OP_NONE, src); } 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)); + wrp_mov(nfp_prog, reg_both(dst), reg_b(src)); +} + +/* wrp_reg_subpart() - load @field_len bytes from @offset of @src, write the + * result to @dst from low end. + */ +static void +wrp_reg_subpart(struct nfp_prog *nfp_prog, swreg dst, swreg src, u8 field_len, + u8 offset) +{ + enum shf_sc sc = offset ? SHF_SC_R_SHF : SHF_SC_NONE; + u8 mask = (1 << field_len) - 1; + + emit_ld_field_any(nfp_prog, dst, mask, src, sc, offset * 8, true); +} + +static void +addr40_offset(struct nfp_prog *nfp_prog, u8 src_gpr, swreg offset, + swreg *rega, swreg *regb) +{ + if (offset == reg_imm(0)) { + *rega = reg_a(src_gpr); + *regb = reg_b(src_gpr + 1); + return; + } + + emit_alu(nfp_prog, imm_a(nfp_prog), reg_a(src_gpr), ALU_OP_ADD, offset); + emit_alu(nfp_prog, imm_b(nfp_prog), reg_b(src_gpr + 1), ALU_OP_ADD_C, + reg_imm(0)); + *rega = imm_a(nfp_prog); + *regb = imm_b(nfp_prog); +} + +/* NFP has Command Push Pull bus which supports bluk memory operations. */ +static int nfp_cpp_memcpy(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + bool descending_seq = meta->ldst_gather_len < 0; + s16 len = abs(meta->ldst_gather_len); + swreg src_base, off; + bool src_40bit_addr; + unsigned int i; + u8 xfer_num; + + off = re_load_imm_any(nfp_prog, meta->insn.off, imm_b(nfp_prog)); + src_40bit_addr = meta->ptr.type == PTR_TO_MAP_VALUE; + src_base = reg_a(meta->insn.src_reg * 2); + xfer_num = round_up(len, 4) / 4; + + if (src_40bit_addr) + addr40_offset(nfp_prog, meta->insn.src_reg, off, &src_base, + &off); + + /* Setup PREV_ALU fields to override memory read length. */ + if (len > 32) + wrp_immed(nfp_prog, reg_none(), + CMD_OVE_LEN | FIELD_PREP(CMD_OV_LEN, xfer_num - 1)); + + /* Memory read from source addr into transfer-in registers. */ + emit_cmd_any(nfp_prog, CMD_TGT_READ32_SWAP, + src_40bit_addr ? CMD_MODE_40b_BA : CMD_MODE_32b, 0, + src_base, off, xfer_num - 1, true, len > 32); + + /* Move from transfer-in to transfer-out. */ + for (i = 0; i < xfer_num; i++) + wrp_mov(nfp_prog, reg_xfer(i), reg_xfer(i)); + + off = re_load_imm_any(nfp_prog, meta->paired_st->off, imm_b(nfp_prog)); + + if (len <= 8) { + /* Use single direct_ref write8. */ + emit_cmd(nfp_prog, CMD_TGT_WRITE8_SWAP, CMD_MODE_32b, 0, + reg_a(meta->paired_st->dst_reg * 2), off, len - 1, + true); + } else if (len <= 32 && IS_ALIGNED(len, 4)) { + /* Use single direct_ref write32. */ + emit_cmd(nfp_prog, CMD_TGT_WRITE32_SWAP, CMD_MODE_32b, 0, + reg_a(meta->paired_st->dst_reg * 2), off, xfer_num - 1, + true); + } else if (len <= 32) { + /* Use single indirect_ref write8. */ + wrp_immed(nfp_prog, reg_none(), + CMD_OVE_LEN | FIELD_PREP(CMD_OV_LEN, len - 1)); + emit_cmd_indir(nfp_prog, CMD_TGT_WRITE8_SWAP, CMD_MODE_32b, 0, + reg_a(meta->paired_st->dst_reg * 2), off, + len - 1, true); + } else if (IS_ALIGNED(len, 4)) { + /* Use single indirect_ref write32. */ + wrp_immed(nfp_prog, reg_none(), + CMD_OVE_LEN | FIELD_PREP(CMD_OV_LEN, xfer_num - 1)); + emit_cmd_indir(nfp_prog, CMD_TGT_WRITE32_SWAP, CMD_MODE_32b, 0, + reg_a(meta->paired_st->dst_reg * 2), off, + xfer_num - 1, true); + } else if (len <= 40) { + /* Use one direct_ref write32 to write the first 32-bytes, then + * another direct_ref write8 to write the remaining bytes. + */ + emit_cmd(nfp_prog, CMD_TGT_WRITE32_SWAP, CMD_MODE_32b, 0, + reg_a(meta->paired_st->dst_reg * 2), off, 7, + true); + + off = re_load_imm_any(nfp_prog, meta->paired_st->off + 32, + imm_b(nfp_prog)); + emit_cmd(nfp_prog, CMD_TGT_WRITE8_SWAP, CMD_MODE_32b, 8, + reg_a(meta->paired_st->dst_reg * 2), off, len - 33, + true); + } else { + /* Use one indirect_ref write32 to write 4-bytes aligned length, + * then another direct_ref write8 to write the remaining bytes. + */ + u8 new_off; + + wrp_immed(nfp_prog, reg_none(), + CMD_OVE_LEN | FIELD_PREP(CMD_OV_LEN, xfer_num - 2)); + emit_cmd_indir(nfp_prog, CMD_TGT_WRITE32_SWAP, CMD_MODE_32b, 0, + reg_a(meta->paired_st->dst_reg * 2), off, + xfer_num - 2, true); + new_off = meta->paired_st->off + (xfer_num - 1) * 4; + off = re_load_imm_any(nfp_prog, new_off, imm_b(nfp_prog)); + emit_cmd(nfp_prog, CMD_TGT_WRITE8_SWAP, CMD_MODE_32b, + xfer_num - 1, reg_a(meta->paired_st->dst_reg * 2), off, + (len & 0x3) - 1, true); + } + + /* TODO: The following extra load is to make sure data flow be identical + * before and after we do memory copy optimization. + * + * The load destination register is not guaranteed to be dead, so we + * need to make sure it is loaded with the value the same as before + * this transformation. + * + * These extra loads could be removed once we have accurate register + * usage information. + */ + if (descending_seq) + xfer_num = 0; + else if (BPF_SIZE(meta->insn.code) != BPF_DW) + xfer_num = xfer_num - 1; + else + xfer_num = xfer_num - 2; + + switch (BPF_SIZE(meta->insn.code)) { + case BPF_B: + wrp_reg_subpart(nfp_prog, reg_both(meta->insn.dst_reg * 2), + reg_xfer(xfer_num), 1, + IS_ALIGNED(len, 4) ? 3 : (len & 3) - 1); + break; + case BPF_H: + wrp_reg_subpart(nfp_prog, reg_both(meta->insn.dst_reg * 2), + reg_xfer(xfer_num), 2, (len & 3) ^ 2); + break; + case BPF_W: + wrp_mov(nfp_prog, reg_both(meta->insn.dst_reg * 2), + reg_xfer(0)); + break; + case BPF_DW: + wrp_mov(nfp_prog, reg_both(meta->insn.dst_reg * 2), + reg_xfer(xfer_num)); + wrp_mov(nfp_prog, reg_both(meta->insn.dst_reg * 2 + 1), + reg_xfer(xfer_num + 1)); + break; + } + + if (BPF_SIZE(meta->insn.code) != BPF_DW) + wrp_immed(nfp_prog, reg_both(meta->insn.dst_reg * 2 + 1), 0); + + return 0; } static int -construct_data_ind_ld(struct nfp_prog *nfp_prog, u16 offset, - u16 src, bool src_valid, u8 size) +data_ld(struct nfp_prog *nfp_prog, swreg offset, u8 dst_gpr, int 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; + sz = max(size, 4); 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); - } + emit_cmd(nfp_prog, CMD_TGT_READ8, CMD_MODE_32b, 0, + pptr_reg(nfp_prog), offset, sz - 1, true); i = 0; if (shift) - emit_shf(nfp_prog, reg_both(0), reg_none(), SHF_OP_NONE, + emit_shf(nfp_prog, reg_both(dst_gpr), 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)); + wrp_mov(nfp_prog, reg_both(dst_gpr + i), reg_xfer(i)); if (i < 2) - wrp_immed(nfp_prog, reg_both(1), 0); + wrp_immed(nfp_prog, reg_both(dst_gpr + 1), 0); return 0; } +static int +data_ld_host_order(struct nfp_prog *nfp_prog, u8 dst_gpr, + swreg lreg, swreg rreg, int size, enum cmd_mode mode) +{ + unsigned int i; + u8 mask, sz; + + /* We load the value from the address indicated in rreg + lreg and then + * mask out the data we don't need. Note: this is little endian! + */ + sz = max(size, 4); + mask = size < 4 ? GENMASK(size - 1, 0) : 0; + + emit_cmd(nfp_prog, CMD_TGT_READ32_SWAP, mode, 0, + lreg, rreg, sz / 4 - 1, true); + + i = 0; + if (mask) + emit_ld_field_any(nfp_prog, reg_both(dst_gpr), mask, + reg_xfer(0), SHF_SC_NONE, 0, true); + else + for (; i * 4 < size; i++) + wrp_mov(nfp_prog, reg_both(dst_gpr + i), reg_xfer(i)); + + if (i < 2) + wrp_immed(nfp_prog, reg_both(dst_gpr + 1), 0); + + return 0; +} + +static int +data_ld_host_order_addr32(struct nfp_prog *nfp_prog, u8 src_gpr, swreg offset, + u8 dst_gpr, u8 size) +{ + return data_ld_host_order(nfp_prog, dst_gpr, reg_a(src_gpr), offset, + size, CMD_MODE_32b); +} + +static int +data_ld_host_order_addr40(struct nfp_prog *nfp_prog, u8 src_gpr, swreg offset, + u8 dst_gpr, u8 size) +{ + swreg rega, regb; + + addr40_offset(nfp_prog, src_gpr, offset, ®a, ®b); + + return data_ld_host_order(nfp_prog, dst_gpr, rega, regb, + size, CMD_MODE_40b_BA); +} + +static int +construct_data_ind_ld(struct nfp_prog *nfp_prog, u16 offset, u16 src, u8 size) +{ + swreg tmp_reg; + + /* 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(), + plen_reg(nfp_prog), ALU_OP_SUB, imm_a(nfp_prog)); + emit_br_relo(nfp_prog, BR_BLO, BR_OFF_RELO, 0, RELO_BR_GO_ABORT); + + /* Load data */ + return data_ld(nfp_prog, imm_b(nfp_prog), 0, size); +} + 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); + swreg tmp_reg; + + /* Check packet length */ + tmp_reg = ur_load_imm_any(nfp_prog, offset + size, imm_a(nfp_prog)); + emit_alu(nfp_prog, reg_none(), plen_reg(nfp_prog), ALU_OP_SUB, tmp_reg); + emit_br_relo(nfp_prog, BR_BLO, BR_OFF_RELO, 0, RELO_BR_GO_ABORT); + + /* Load data */ + tmp_reg = re_load_imm_any(nfp_prog, offset, imm_b(nfp_prog)); + return data_ld(nfp_prog, tmp_reg, 0, size); } -static int wrp_set_mark(struct nfp_prog *nfp_prog, u8 src) +static int +data_stx_host_order(struct nfp_prog *nfp_prog, u8 dst_gpr, swreg offset, + u8 src_gpr, u8 size) { - 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)); + unsigned int i; + + for (i = 0; i * 4 < size; i++) + wrp_mov(nfp_prog, reg_xfer(i), reg_a(src_gpr + i)); + + emit_cmd(nfp_prog, CMD_TGT_WRITE8_SWAP, CMD_MODE_32b, 0, + reg_a(dst_gpr), offset, size - 1, true); + + return 0; +} + +static int +data_st_host_order(struct nfp_prog *nfp_prog, u8 dst_gpr, swreg offset, + u64 imm, u8 size) +{ + wrp_immed(nfp_prog, reg_xfer(0), imm); + if (size == 8) + wrp_immed(nfp_prog, reg_xfer(1), imm >> 32); + + emit_cmd(nfp_prog, CMD_TGT_WRITE8_SWAP, CMD_MODE_32b, 0, + reg_a(dst_gpr), offset, size - 1, true); + + return 0; +} + +typedef int +(*lmem_step)(struct nfp_prog *nfp_prog, u8 gpr, u8 gpr_byte, s32 off, + unsigned int size, bool first, bool new_gpr, bool last, bool lm3, + bool needs_inc); + +static int +wrp_lmem_load(struct nfp_prog *nfp_prog, u8 dst, u8 dst_byte, s32 off, + unsigned int size, bool first, bool new_gpr, bool last, bool lm3, + bool needs_inc) +{ + bool should_inc = needs_inc && new_gpr && !last; + u32 idx, src_byte; + enum shf_sc sc; + swreg reg; + int shf; + u8 mask; + + if (WARN_ON_ONCE(dst_byte + size > 4 || off % 4 + size > 4)) + return -EOPNOTSUPP; + + idx = off / 4; + + /* Move the entire word */ + if (size == 4) { + wrp_mov(nfp_prog, reg_both(dst), + should_inc ? reg_lm_inc(3) : reg_lm(lm3 ? 3 : 0, idx)); + return 0; + } + + if (WARN_ON_ONCE(lm3 && idx > RE_REG_LM_IDX_MAX)) + return -EOPNOTSUPP; + + src_byte = off % 4; + + mask = (1 << size) - 1; + mask <<= dst_byte; + + if (WARN_ON_ONCE(mask > 0xf)) + return -EOPNOTSUPP; + + shf = abs(src_byte - dst_byte) * 8; + if (src_byte == dst_byte) { + sc = SHF_SC_NONE; + } else if (src_byte < dst_byte) { + shf = 32 - shf; + sc = SHF_SC_L_SHF; + } else { + sc = SHF_SC_R_SHF; + } + + /* ld_field can address fewer indexes, if offset too large do RMW. + * Because we RMV twice we waste 2 cycles on unaligned 8 byte writes. + */ + if (idx <= RE_REG_LM_IDX_MAX) { + reg = reg_lm(lm3 ? 3 : 0, idx); + } else { + reg = imm_a(nfp_prog); + /* If it's not the first part of the load and we start a new GPR + * that means we are loading a second part of the LMEM word into + * a new GPR. IOW we've already looked that LMEM word and + * therefore it has been loaded into imm_a(). + */ + if (first || !new_gpr) + wrp_mov(nfp_prog, reg, reg_lm(0, idx)); + } + + emit_ld_field_any(nfp_prog, reg_both(dst), mask, reg, sc, shf, new_gpr); + + if (should_inc) + wrp_mov(nfp_prog, reg_none(), reg_lm_inc(3)); + + return 0; +} + +static int +wrp_lmem_store(struct nfp_prog *nfp_prog, u8 src, u8 src_byte, s32 off, + unsigned int size, bool first, bool new_gpr, bool last, bool lm3, + bool needs_inc) +{ + bool should_inc = needs_inc && new_gpr && !last; + u32 idx, dst_byte; + enum shf_sc sc; + swreg reg; + int shf; + u8 mask; + + if (WARN_ON_ONCE(src_byte + size > 4 || off % 4 + size > 4)) + return -EOPNOTSUPP; + + idx = off / 4; + + /* Move the entire word */ + if (size == 4) { + wrp_mov(nfp_prog, + should_inc ? reg_lm_inc(3) : reg_lm(lm3 ? 3 : 0, idx), + reg_b(src)); + return 0; + } + + if (WARN_ON_ONCE(lm3 && idx > RE_REG_LM_IDX_MAX)) + return -EOPNOTSUPP; + + dst_byte = off % 4; + + mask = (1 << size) - 1; + mask <<= dst_byte; + + if (WARN_ON_ONCE(mask > 0xf)) + return -EOPNOTSUPP; + + shf = abs(src_byte - dst_byte) * 8; + if (src_byte == dst_byte) { + sc = SHF_SC_NONE; + } else if (src_byte < dst_byte) { + shf = 32 - shf; + sc = SHF_SC_L_SHF; + } else { + sc = SHF_SC_R_SHF; + } + + /* ld_field can address fewer indexes, if offset too large do RMW. + * Because we RMV twice we waste 2 cycles on unaligned 8 byte writes. + */ + if (idx <= RE_REG_LM_IDX_MAX) { + reg = reg_lm(lm3 ? 3 : 0, idx); + } else { + reg = imm_a(nfp_prog); + /* Only first and last LMEM locations are going to need RMW, + * the middle location will be overwritten fully. + */ + if (first || last) + wrp_mov(nfp_prog, reg, reg_lm(0, idx)); + } + + emit_ld_field(nfp_prog, reg, mask, reg_b(src), sc, shf); + + if (new_gpr || last) { + if (idx > RE_REG_LM_IDX_MAX) + wrp_mov(nfp_prog, reg_lm(0, idx), reg); + if (should_inc) + wrp_mov(nfp_prog, reg_none(), reg_lm_inc(3)); + } + + return 0; +} + +static int +mem_op_stack(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, + unsigned int size, unsigned int ptr_off, u8 gpr, u8 ptr_gpr, + bool clr_gpr, lmem_step step) +{ + s32 off = nfp_prog->stack_depth + meta->insn.off + ptr_off; + bool first = true, last; + bool needs_inc = false; + swreg stack_off_reg; + u8 prev_gpr = 255; + u32 gpr_byte = 0; + bool lm3 = true; + int ret; + + if (meta->ptr_not_const) { + /* Use of the last encountered ptr_off is OK, they all have + * the same alignment. Depend on low bits of value being + * discarded when written to LMaddr register. + */ + stack_off_reg = ur_load_imm_any(nfp_prog, meta->insn.off, + stack_imm(nfp_prog)); + + emit_alu(nfp_prog, imm_b(nfp_prog), + reg_a(ptr_gpr), ALU_OP_ADD, stack_off_reg); + + needs_inc = true; + } else if (off + size <= 64) { + /* We can reach bottom 64B with LMaddr0 */ + lm3 = false; + } else if (round_down(off, 32) == round_down(off + size - 1, 32)) { + /* We have to set up a new pointer. If we know the offset + * and the entire access falls into a single 32 byte aligned + * window we won't have to increment the LM pointer. + * The 32 byte alignment is imporant because offset is ORed in + * not added when doing *l$indexN[off]. + */ + stack_off_reg = ur_load_imm_any(nfp_prog, round_down(off, 32), + stack_imm(nfp_prog)); + emit_alu(nfp_prog, imm_b(nfp_prog), + stack_reg(nfp_prog), ALU_OP_ADD, stack_off_reg); + + off %= 32; + } else { + stack_off_reg = ur_load_imm_any(nfp_prog, round_down(off, 4), + stack_imm(nfp_prog)); + + emit_alu(nfp_prog, imm_b(nfp_prog), + stack_reg(nfp_prog), ALU_OP_ADD, stack_off_reg); + + needs_inc = true; + } + if (lm3) { + emit_csr_wr(nfp_prog, imm_b(nfp_prog), NFP_CSR_ACT_LM_ADDR3); + /* For size < 4 one slot will be filled by zeroing of upper. */ + wrp_nops(nfp_prog, clr_gpr && size < 8 ? 2 : 3); + } + + if (clr_gpr && size < 8) + wrp_immed(nfp_prog, reg_both(gpr + 1), 0); + + while (size) { + u32 slice_end; + u8 slice_size; + + slice_size = min(size, 4 - gpr_byte); + slice_end = min(off + slice_size, round_up(off + 1, 4)); + slice_size = slice_end - off; + + last = slice_size == size; + + if (needs_inc) + off %= 4; + + ret = step(nfp_prog, gpr, gpr_byte, off, slice_size, + first, gpr != prev_gpr, last, lm3, needs_inc); + if (ret) + return ret; + + prev_gpr = gpr; + first = false; + + gpr_byte += slice_size; + if (gpr_byte >= 4) { + gpr_byte -= 4; + gpr++; + } + + size -= slice_size; + off += slice_size; + } return 0; } @@ -697,7 +1090,7 @@ static int wrp_set_mark(struct nfp_prog *nfp_prog, u8 src) static void wrp_alu_imm(struct nfp_prog *nfp_prog, u8 dst, enum alu_op alu_op, u32 imm) { - u32 tmp_reg; + swreg tmp_reg; if (alu_op == ALU_OP_AND) { if (!imm) @@ -714,7 +1107,7 @@ wrp_alu_imm(struct nfp_prog *nfp_prog, u8 dst, enum alu_op alu_op, u32 imm) if (alu_op == ALU_OP_XOR) { if (!~imm) emit_alu(nfp_prog, reg_both(dst), reg_none(), - ALU_OP_NEG, reg_b(dst)); + ALU_OP_NOT, reg_b(dst)); if (!imm || !~imm) return; } @@ -797,9 +1190,6 @@ wrp_test_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, { const struct bpf_insn *insn = &meta->insn; - if (insn->off < 0) /* TODO */ - return -EOPNOTSUPP; - 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, @@ -815,10 +1205,7 @@ wrp_cmp_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, 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 -EOPNOTSUPP; + swreg tmp_reg; tmp_reg = ur_load_imm_any(nfp_prog, imm & ~0U, imm_b(nfp_prog)); if (!swap) @@ -844,10 +1231,10 @@ 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; + u8 areg, breg; - if (insn->off < 0) /* TODO */ - return -EOPNOTSUPP; + areg = insn->dst_reg * 2; + breg = insn->src_reg * 2; if (swap) { areg ^= breg; @@ -863,13 +1250,164 @@ wrp_cmp_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, return 0; } +static void wrp_end32(struct nfp_prog *nfp_prog, swreg reg_in, u8 gpr_out) +{ + emit_ld_field(nfp_prog, reg_both(gpr_out), 0xf, reg_in, + SHF_SC_R_ROT, 8); + emit_ld_field(nfp_prog, reg_both(gpr_out), 0x5, reg_a(gpr_out), + SHF_SC_R_ROT, 16); +} + +static int adjust_head(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + swreg tmp = imm_a(nfp_prog), tmp_len = imm_b(nfp_prog); + struct nfp_bpf_cap_adjust_head *adjust_head; + u32 ret_einval, end; + + adjust_head = &nfp_prog->bpf->adjust_head; + + /* Optimized version - 5 vs 14 cycles */ + if (nfp_prog->adjust_head_location != UINT_MAX) { + if (WARN_ON_ONCE(nfp_prog->adjust_head_location != meta->n)) + return -EINVAL; + + emit_alu(nfp_prog, pptr_reg(nfp_prog), + reg_a(2 * 2), ALU_OP_ADD, pptr_reg(nfp_prog)); + emit_alu(nfp_prog, plen_reg(nfp_prog), + plen_reg(nfp_prog), ALU_OP_SUB, reg_a(2 * 2)); + emit_alu(nfp_prog, pv_len(nfp_prog), + pv_len(nfp_prog), ALU_OP_SUB, reg_a(2 * 2)); + + wrp_immed(nfp_prog, reg_both(0), 0); + wrp_immed(nfp_prog, reg_both(1), 0); + + /* TODO: when adjust head is guaranteed to succeed we can + * also eliminate the following if (r0 == 0) branch. + */ + + return 0; + } + + ret_einval = nfp_prog_current_offset(nfp_prog) + 14; + end = ret_einval + 2; + + /* We need to use a temp because offset is just a part of the pkt ptr */ + emit_alu(nfp_prog, tmp, + reg_a(2 * 2), ALU_OP_ADD_2B, pptr_reg(nfp_prog)); + + /* Validate result will fit within FW datapath constraints */ + emit_alu(nfp_prog, reg_none(), + tmp, ALU_OP_SUB, reg_imm(adjust_head->off_min)); + emit_br(nfp_prog, BR_BLO, ret_einval, 0); + emit_alu(nfp_prog, reg_none(), + reg_imm(adjust_head->off_max), ALU_OP_SUB, tmp); + emit_br(nfp_prog, BR_BLO, ret_einval, 0); + + /* Validate the length is at least ETH_HLEN */ + emit_alu(nfp_prog, tmp_len, + plen_reg(nfp_prog), ALU_OP_SUB, reg_a(2 * 2)); + emit_alu(nfp_prog, reg_none(), + tmp_len, ALU_OP_SUB, reg_imm(ETH_HLEN)); + emit_br(nfp_prog, BR_BMI, ret_einval, 0); + + /* Load the ret code */ + wrp_immed(nfp_prog, reg_both(0), 0); + wrp_immed(nfp_prog, reg_both(1), 0); + + /* Modify the packet metadata */ + emit_ld_field(nfp_prog, pptr_reg(nfp_prog), 0x3, tmp, SHF_SC_NONE, 0); + + /* Skip over the -EINVAL ret code (defer 2) */ + emit_br(nfp_prog, BR_UNC, end, 2); + + emit_alu(nfp_prog, plen_reg(nfp_prog), + plen_reg(nfp_prog), ALU_OP_SUB, reg_a(2 * 2)); + emit_alu(nfp_prog, pv_len(nfp_prog), + pv_len(nfp_prog), ALU_OP_SUB, reg_a(2 * 2)); + + /* return -EINVAL target */ + if (!nfp_prog_confirm_current_offset(nfp_prog, ret_einval)) + return -EINVAL; + + wrp_immed(nfp_prog, reg_both(0), -22); + wrp_immed(nfp_prog, reg_both(1), ~0); + + if (!nfp_prog_confirm_current_offset(nfp_prog, end)) + return -EINVAL; + + return 0; +} + +static int +map_lookup_stack(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; + lm_off += meta->arg2.var_off.value + meta->arg2.off; + load_lm_ptr = meta->arg2_var_off || lm_off; + + /* Set LM0 to start of key */ + if (load_lm_ptr) + emit_csr_wr(nfp_prog, reg_b(2 * 2), NFP_CSR_ACT_LM_ADDR0); + + /* 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 + BPF_FUNC_map_lookup_elem, + 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); + + /* 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; + + /* Reset the LM0 pointer */ + if (!load_lm_ptr) + return 0; + + emit_csr_wr(nfp_prog, stack_reg(nfp_prog), NFP_CSR_ACT_LM_ADDR0); + wrp_nops(nfp_prog, 3); + + 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); + u8 dst = insn->dst_reg * 2; + u8 src = insn->src_reg * 2; + + if (insn->src_reg == BPF_REG_10) { + swreg stack_depth_reg; + + stack_depth_reg = ur_load_imm_any(nfp_prog, + nfp_prog->stack_depth, + stack_imm(nfp_prog)); + emit_alu(nfp_prog, reg_both(dst), + stack_reg(nfp_prog), ALU_OP_ADD, stack_depth_reg); + wrp_immed(nfp_prog, reg_both(dst + 1), 0); + } else { + wrp_reg_mov(nfp_prog, dst, src); + wrp_reg_mov(nfp_prog, dst + 1, src + 1); + } return 0; } @@ -964,28 +1502,64 @@ static int sub_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) return 0; } -static int shl_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +static int neg_reg64(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); + emit_alu(nfp_prog, reg_both(insn->dst_reg * 2), reg_imm(0), + ALU_OP_SUB, reg_b(insn->dst_reg * 2)); + emit_alu(nfp_prog, reg_both(insn->dst_reg * 2 + 1), reg_imm(0), + ALU_OP_SUB_C, reg_b(insn->dst_reg * 2 + 1)); return 0; } -static int shr_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +static int shl_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { const struct bpf_insn *insn = &meta->insn; + u8 dst = insn->dst_reg * 2; + + if (insn->imm < 32) { + emit_shf(nfp_prog, reg_both(dst + 1), + reg_a(dst + 1), SHF_OP_NONE, reg_b(dst), + SHF_SC_R_DSHF, 32 - insn->imm); + emit_shf(nfp_prog, reg_both(dst), + reg_none(), SHF_OP_NONE, reg_b(dst), + SHF_SC_L_SHF, insn->imm); + } else if (insn->imm == 32) { + wrp_reg_mov(nfp_prog, dst + 1, dst); + wrp_immed(nfp_prog, reg_both(dst), 0); + } else if (insn->imm > 32) { + emit_shf(nfp_prog, reg_both(dst + 1), + reg_none(), SHF_OP_NONE, reg_b(dst), + SHF_SC_L_SHF, insn->imm - 32); + wrp_immed(nfp_prog, reg_both(dst), 0); + } - if (insn->imm != 32) - return 1; /* TODO */ + return 0; +} - 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); +static int shr_imm64(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + const struct bpf_insn *insn = &meta->insn; + u8 dst = insn->dst_reg * 2; + + if (insn->imm < 32) { + emit_shf(nfp_prog, reg_both(dst), + reg_a(dst + 1), SHF_OP_NONE, reg_b(dst), + SHF_SC_R_DSHF, insn->imm); + emit_shf(nfp_prog, reg_both(dst + 1), + reg_none(), SHF_OP_NONE, reg_b(dst + 1), + SHF_SC_R_SHF, insn->imm); + } else if (insn->imm == 32) { + wrp_reg_mov(nfp_prog, dst, dst + 1); + wrp_immed(nfp_prog, reg_both(dst + 1), 0); + } else if (insn->imm > 32) { + emit_shf(nfp_prog, reg_both(dst), + reg_none(), SHF_OP_NONE, reg_b(dst + 1), + SHF_SC_R_SHF, insn->imm - 32); + wrp_immed(nfp_prog, reg_both(dst + 1), 0); + } return 0; } @@ -1060,6 +1634,16 @@ 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 neg_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + u8 dst = meta->insn.dst_reg * 2; + + emit_alu(nfp_prog, reg_both(dst), reg_imm(0), ALU_OP_SUB, reg_b(dst)); + wrp_immed(nfp_prog, reg_both(meta->insn.dst_reg * 2 + 1), 0); + + return 0; +} + static int shl_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { const struct bpf_insn *insn = &meta->insn; @@ -1075,21 +1659,59 @@ static int shl_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) return 0; } +static int end_reg32(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + const struct bpf_insn *insn = &meta->insn; + u8 gpr = insn->dst_reg * 2; + + switch (insn->imm) { + case 16: + emit_ld_field(nfp_prog, reg_both(gpr), 0x9, reg_b(gpr), + SHF_SC_R_ROT, 8); + emit_ld_field(nfp_prog, reg_both(gpr), 0xe, reg_a(gpr), + SHF_SC_R_SHF, 16); + + wrp_immed(nfp_prog, reg_both(gpr + 1), 0); + break; + case 32: + wrp_end32(nfp_prog, reg_a(gpr), gpr); + wrp_immed(nfp_prog, reg_both(gpr + 1), 0); + break; + case 64: + wrp_mov(nfp_prog, imm_a(nfp_prog), reg_b(gpr + 1)); + + wrp_end32(nfp_prog, reg_a(gpr), gpr + 1); + wrp_end32(nfp_prog, imm_a(nfp_prog), gpr); + break; + } + + 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); + struct nfp_insn_meta *prev = nfp_meta_prev(meta); + u32 imm_lo, imm_hi; + u8 dst; + + dst = prev->insn.dst_reg * 2; + imm_lo = prev->insn.imm; + imm_hi = meta->insn.imm; + + wrp_immed(nfp_prog, reg_both(dst), imm_lo); + + /* mov is always 1 insn, load imm may be two, so try to use mov */ + if (imm_hi == imm_lo) + wrp_mov(nfp_prog, reg_both(dst + 1), reg_a(dst)); + else + wrp_immed(nfp_prog, reg_both(dst + 1), imm_hi); 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; } @@ -1111,88 +1733,257 @@ static int data_ld4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) 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); + meta->insn.src_reg * 2, 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); + meta->insn.src_reg * 2, 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); + meta->insn.src_reg * 2, 4); } -static int mem_ldx4_skb(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +static int +mem_ldx_stack(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, + unsigned int size, unsigned int ptr_off) { - 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 mem_op_stack(nfp_prog, meta, size, ptr_off, + meta->insn.dst_reg * 2, meta->insn.src_reg * 2, + true, wrp_lmem_load); +} + +static int mem_ldx_skb(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, + u8 size) +{ + swreg dst = reg_both(meta->insn.dst_reg * 2); + + switch (meta->insn.off) { + case offsetof(struct __sk_buff, len): + if (size != FIELD_SIZEOF(struct __sk_buff, len)) + return -EOPNOTSUPP; + wrp_mov(nfp_prog, dst, plen_reg(nfp_prog)); + break; + case offsetof(struct __sk_buff, data): + if (size != FIELD_SIZEOF(struct __sk_buff, data)) + return -EOPNOTSUPP; + wrp_mov(nfp_prog, dst, pptr_reg(nfp_prog)); + break; + case offsetof(struct __sk_buff, data_end): + if (size != FIELD_SIZEOF(struct __sk_buff, data_end)) + return -EOPNOTSUPP; + emit_alu(nfp_prog, dst, + plen_reg(nfp_prog), ALU_OP_ADD, pptr_reg(nfp_prog)); + break; + default: return -EOPNOTSUPP; + } + + wrp_immed(nfp_prog, reg_both(meta->insn.dst_reg * 2 + 1), 0); return 0; } -static int mem_ldx4_xdp(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +static int mem_ldx_xdp(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, + u8 size) { - u32 dst = reg_both(meta->insn.dst_reg * 2); + swreg dst = reg_both(meta->insn.dst_reg * 2); - if (meta->insn.off != offsetof(struct xdp_md, data) && - meta->insn.off != offsetof(struct xdp_md, data_end)) + switch (meta->insn.off) { + case offsetof(struct xdp_md, data): + if (size != FIELD_SIZEOF(struct xdp_md, data)) + return -EOPNOTSUPP; + wrp_mov(nfp_prog, dst, pptr_reg(nfp_prog)); + break; + case offsetof(struct xdp_md, data_end): + if (size != FIELD_SIZEOF(struct xdp_md, data_end)) + return -EOPNOTSUPP; + emit_alu(nfp_prog, dst, + plen_reg(nfp_prog), ALU_OP_ADD, pptr_reg(nfp_prog)); + break; + default: return -EOPNOTSUPP; + } - emit_alu(nfp_prog, dst, reg_none(), ALU_OP_NONE, NFP_BPF_ABI_PKT); + wrp_immed(nfp_prog, reg_both(meta->insn.dst_reg * 2 + 1), 0); - if (meta->insn.off == offsetof(struct xdp_md, data)) - return 0; + return 0; +} - emit_alu(nfp_prog, dst, dst, ALU_OP_ADD, NFP_BPF_ABI_LEN); +static int +mem_ldx_data(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, + unsigned int size) +{ + swreg tmp_reg; - return 0; + tmp_reg = re_load_imm_any(nfp_prog, meta->insn.off, imm_b(nfp_prog)); + + return data_ld_host_order_addr32(nfp_prog, meta->insn.src_reg * 2, + tmp_reg, meta->insn.dst_reg * 2, size); +} + +static int +mem_ldx_emem(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, + unsigned int size) +{ + swreg tmp_reg; + + tmp_reg = re_load_imm_any(nfp_prog, meta->insn.off, imm_b(nfp_prog)); + + return data_ld_host_order_addr40(nfp_prog, meta->insn.src_reg * 2, + tmp_reg, meta->insn.dst_reg * 2, size); +} + +static int +mem_ldx(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, + unsigned int size) +{ + if (meta->ldst_gather_len) + return nfp_cpp_memcpy(nfp_prog, meta); + + if (meta->ptr.type == PTR_TO_CTX) { + if (nfp_prog->type == BPF_PROG_TYPE_XDP) + return mem_ldx_xdp(nfp_prog, meta, size); + else + return mem_ldx_skb(nfp_prog, meta, size); + } + + if (meta->ptr.type == PTR_TO_PACKET) + return mem_ldx_data(nfp_prog, meta, size); + + if (meta->ptr.type == PTR_TO_STACK) + return mem_ldx_stack(nfp_prog, meta, size, + meta->ptr.off + meta->ptr.var_off.value); + + if (meta->ptr.type == PTR_TO_MAP_VALUE) + return mem_ldx_emem(nfp_prog, meta, size); + + return -EOPNOTSUPP; +} + +static int mem_ldx1(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return mem_ldx(nfp_prog, meta, 1); +} + +static int mem_ldx2(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return mem_ldx(nfp_prog, meta, 2); } static int mem_ldx4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { - int ret; + return mem_ldx(nfp_prog, meta, 4); +} - if (nfp_prog->act == NN_ACT_XDP) - ret = mem_ldx4_xdp(nfp_prog, meta); - else - ret = mem_ldx4_skb(nfp_prog, meta); +static int mem_ldx8(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return mem_ldx(nfp_prog, meta, 8); +} - wrp_immed(nfp_prog, reg_both(meta->insn.dst_reg * 2 + 1), 0); +static int +mem_st_data(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, + unsigned int size) +{ + u64 imm = meta->insn.imm; /* sign extend */ + swreg off_reg; - return ret; + off_reg = re_load_imm_any(nfp_prog, meta->insn.off, imm_b(nfp_prog)); + + return data_st_host_order(nfp_prog, meta->insn.dst_reg * 2, off_reg, + imm, size); } -static int mem_stx4_skb(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +static int mem_st(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, + unsigned int size) { - if (meta->insn.off == offsetof(struct sk_buff, mark)) - return wrp_set_mark(nfp_prog, meta->insn.src_reg * 2); + if (meta->ptr.type == PTR_TO_PACKET) + return mem_st_data(nfp_prog, meta, size); return -EOPNOTSUPP; } -static int mem_stx4_xdp(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +static int mem_st1(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return mem_st(nfp_prog, meta, 1); +} + +static int mem_st2(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { + return mem_st(nfp_prog, meta, 2); +} + +static int mem_st4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return mem_st(nfp_prog, meta, 4); +} + +static int mem_st8(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return mem_st(nfp_prog, meta, 8); +} + +static int +mem_stx_data(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, + unsigned int size) +{ + swreg off_reg; + + off_reg = re_load_imm_any(nfp_prog, meta->insn.off, imm_b(nfp_prog)); + + return data_stx_host_order(nfp_prog, meta->insn.dst_reg * 2, off_reg, + meta->insn.src_reg * 2, size); +} + +static int +mem_stx_stack(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, + unsigned int size, unsigned int ptr_off) +{ + return mem_op_stack(nfp_prog, meta, size, ptr_off, + meta->insn.src_reg * 2, meta->insn.dst_reg * 2, + false, wrp_lmem_store); +} + +static int +mem_stx(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta, + unsigned int size) +{ + if (meta->ptr.type == PTR_TO_PACKET) + return mem_stx_data(nfp_prog, meta, size); + + if (meta->ptr.type == PTR_TO_STACK) + return mem_stx_stack(nfp_prog, meta, size, + meta->ptr.off + meta->ptr.var_off.value); + return -EOPNOTSUPP; } +static int mem_stx1(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return mem_stx(nfp_prog, meta, 1); +} + +static int mem_stx2(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return mem_stx(nfp_prog, meta, 2); +} + static int mem_stx4(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { - if (nfp_prog->act == NN_ACT_XDP) - return mem_stx4_xdp(nfp_prog, meta); - return mem_stx4_skb(nfp_prog, meta); + return mem_stx(nfp_prog, meta, 4); +} + +static int mem_stx8(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return mem_stx(nfp_prog, meta, 8); } static int jump(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { - if (meta->insn.off < 0) /* TODO */ - return -EOPNOTSUPP; emit_br(nfp_prog, BR_UNC, meta->insn.off, 0); return 0; @@ -1202,11 +1993,10 @@ 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; + swreg or1, or2, tmp_reg; - if (insn->off < 0) /* TODO */ - return -EOPNOTSUPP; + or1 = reg_a(insn->dst_reg * 2); + or2 = reg_b(insn->dst_reg * 2 + 1); if (imm & ~0U) { tmp_reg = ur_load_imm_any(nfp_prog, imm & ~0U, imm_b(nfp_prog)); @@ -1230,32 +2020,49 @@ static int jeq_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) static int jgt_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { - return wrp_cmp_imm(nfp_prog, meta, BR_BLO, false); + return wrp_cmp_imm(nfp_prog, meta, BR_BLO, true); } static int jge_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { - return wrp_cmp_imm(nfp_prog, meta, BR_BHS, true); + return wrp_cmp_imm(nfp_prog, meta, BR_BHS, false); } static int jlt_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { - return wrp_cmp_imm(nfp_prog, meta, BR_BHS, false); + return wrp_cmp_imm(nfp_prog, meta, BR_BLO, false); } static int jle_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { - return wrp_cmp_imm(nfp_prog, meta, BR_BLO, true); + return wrp_cmp_imm(nfp_prog, meta, BR_BHS, true); +} + +static int jsgt_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return wrp_cmp_imm(nfp_prog, meta, BR_BLT, true); +} + +static int jsge_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return wrp_cmp_imm(nfp_prog, meta, BR_BGE, false); +} + +static int jslt_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return wrp_cmp_imm(nfp_prog, meta, BR_BLT, false); +} + +static int jsle_imm(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return wrp_cmp_imm(nfp_prog, meta, BR_BGE, 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 -EOPNOTSUPP; + swreg tmp_reg; if (!imm) { meta->skip = true; @@ -1283,15 +2090,13 @@ 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 -EOPNOTSUPP; + swreg tmp_reg; 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); + return 0; } tmp_reg = ur_load_imm_any(nfp_prog, imm & ~0U, imm_b(nfp_prog)); @@ -1311,9 +2116,6 @@ 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 -EOPNOTSUPP; - 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), @@ -1327,22 +2129,42 @@ static int jeq_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) static int jgt_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { - return wrp_cmp_reg(nfp_prog, meta, BR_BLO, false); + return wrp_cmp_reg(nfp_prog, meta, BR_BLO, true); } static int jge_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { - return wrp_cmp_reg(nfp_prog, meta, BR_BHS, true); + return wrp_cmp_reg(nfp_prog, meta, BR_BHS, false); } static int jlt_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { - return wrp_cmp_reg(nfp_prog, meta, BR_BHS, false); + return wrp_cmp_reg(nfp_prog, meta, BR_BLO, false); } static int jle_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) { - return wrp_cmp_reg(nfp_prog, meta, BR_BLO, true); + return wrp_cmp_reg(nfp_prog, meta, BR_BHS, true); +} + +static int jsgt_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return wrp_cmp_reg(nfp_prog, meta, BR_BLT, true); +} + +static int jsge_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return wrp_cmp_reg(nfp_prog, meta, BR_BGE, false); +} + +static int jslt_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return wrp_cmp_reg(nfp_prog, meta, BR_BLT, false); +} + +static int jsle_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + return wrp_cmp_reg(nfp_prog, meta, BR_BGE, true); } static int jset_reg(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) @@ -1355,9 +2177,22 @@ 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 call(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta) +{ + switch (meta->insn.imm) { + case BPF_FUNC_xdp_adjust_head: + return adjust_head(nfp_prog, meta); + case BPF_FUNC_map_lookup_elem: + return map_lookup_stack(nfp_prog, meta); + default: + WARN_ONCE(1, "verifier allowed unsupported function\n"); + return -EOPNOTSUPP; + } +} + 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); + emit_br_relo(nfp_prog, BR_UNC, BR_OFF_RELO, 0, RELO_BR_GO_OUT); return 0; } @@ -1375,6 +2210,7 @@ static const instr_cb_t instr_cb[256] = { [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_NEG] = neg_reg64, [BPF_ALU64 | BPF_LSH | BPF_K] = shl_imm64, [BPF_ALU64 | BPF_RSH | BPF_K] = shr_imm64, [BPF_ALU | BPF_MOV | BPF_X] = mov_reg, @@ -1389,7 +2225,9 @@ static const instr_cb_t instr_cb[256] = { [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_NEG] = neg_reg, [BPF_ALU | BPF_LSH | BPF_K] = shl_imm, + [BPF_ALU | BPF_END | BPF_X] = end_reg32, [BPF_LD | BPF_IMM | BPF_DW] = imm_ld8, [BPF_LD | BPF_ABS | BPF_B] = data_ld1, [BPF_LD | BPF_ABS | BPF_H] = data_ld2, @@ -1397,14 +2235,28 @@ static const instr_cb_t instr_cb[256] = { [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_B] = mem_ldx1, + [BPF_LDX | BPF_MEM | BPF_H] = mem_ldx2, [BPF_LDX | BPF_MEM | BPF_W] = mem_ldx4, + [BPF_LDX | BPF_MEM | BPF_DW] = mem_ldx8, + [BPF_STX | BPF_MEM | BPF_B] = mem_stx1, + [BPF_STX | BPF_MEM | BPF_H] = mem_stx2, [BPF_STX | BPF_MEM | BPF_W] = mem_stx4, + [BPF_STX | BPF_MEM | BPF_DW] = mem_stx8, + [BPF_ST | BPF_MEM | BPF_B] = mem_st1, + [BPF_ST | BPF_MEM | BPF_H] = mem_st2, + [BPF_ST | BPF_MEM | BPF_W] = mem_st4, + [BPF_ST | BPF_MEM | BPF_DW] = mem_st8, [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_JLT | BPF_K] = jlt_imm, [BPF_JMP | BPF_JLE | BPF_K] = jle_imm, + [BPF_JMP | BPF_JSGT | BPF_K] = jsgt_imm, + [BPF_JMP | BPF_JSGE | BPF_K] = jsge_imm, + [BPF_JMP | BPF_JSLT | BPF_K] = jslt_imm, + [BPF_JMP | BPF_JSLE | BPF_K] = jsle_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, @@ -1412,135 +2264,72 @@ static const instr_cb_t instr_cb[256] = { [BPF_JMP | BPF_JGE | BPF_X] = jge_reg, [BPF_JMP | BPF_JLT | BPF_X] = jlt_reg, [BPF_JMP | BPF_JLE | BPF_X] = jle_reg, + [BPF_JMP | BPF_JSGT | BPF_X] = jsgt_reg, + [BPF_JMP | BPF_JSGE | BPF_X] = jsge_reg, + [BPF_JMP | BPF_JSLT | BPF_X] = jslt_reg, + [BPF_JMP | BPF_JSLE | BPF_X] = jsle_reg, [BPF_JMP | BPF_JSET | BPF_X] = jset_reg, [BPF_JMP | BPF_JNE | BPF_X] = jne_reg, + [BPF_JMP | BPF_CALL] = call, [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; + struct nfp_insn_meta *meta, *jmp_dst; + u32 idx, br_idx; - nfp_for_each_insn_walk2(nfp_prog, meta, next) { + list_for_each_entry(meta, &nfp_prog->insns, l) { if (meta->skip) continue; + if (meta->insn.code == (BPF_JMP | BPF_CALL)) + continue; if (BPF_CLASS(meta->insn.code) != BPF_JMP) continue; - br_idx = nfp_prog_offset_to_index(nfp_prog, next->off) - 1; + if (list_is_last(&meta->l, &nfp_prog->insns)) + br_idx = nfp_prog->last_bpf_off; + else + br_idx = list_next_entry(meta, l)->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])) + if (FIELD_GET(OP_RELO_TYPE, nfp_prog->prog[br_idx]) != + RELO_BR_REL) continue; - /* Find the target offset in assembler realm */ - off = meta->insn.off; - if (!off) { - pr_err("Fixup found zero offset!!\n"); + if (!meta->jmp_dst) { + pr_err("Non-exit jump doesn't have destination info recorded!!\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; - } + jmp_dst = meta->jmp_dst; - if (next->skip) { + if (jmp_dst->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++) { + for (idx = meta->off; idx <= br_idx; idx++) { if (!nfp_is_br(nfp_prog->prog[idx])) continue; - br_set_offset(&nfp_prog->prog[idx], next->off); + br_set_offset(&nfp_prog->prog[idx], jmp_dst->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, - [NN_ACT_TC_REDIR] = 0x24 - }; - /* 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); + wrp_immed(nfp_prog, plen_reg(nfp_prog), GENMASK(13, 0)); + emit_alu(nfp_prog, plen_reg(nfp_prog), + plen_reg(nfp_prog), ALU_OP_AND, pv_len(nfp_prog)); } static void nfp_outro_tc_da(struct nfp_prog *nfp_prog) @@ -1560,10 +2349,9 @@ static void nfp_outro_tc_da(struct nfp_prog *nfp_prog) /* Target for aborts */ nfp_prog->tgt_abort = nfp_prog_current_offset(nfp_prog); - emit_br_def(nfp_prog, nfp_prog->tgt_done, 2); + emit_br_relo(nfp_prog, BR_UNC, BR_OFF_RELO, 2, RELO_BR_NEXT_PKT); - emit_alu(nfp_prog, reg_a(0), - reg_none(), ALU_OP_NONE, NFP_BPF_ABI_FLAGS); + wrp_mov(nfp_prog, reg_a(0), 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 */ @@ -1572,8 +2360,7 @@ static void nfp_outro_tc_da(struct nfp_prog *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_mov(nfp_prog, reg_a(0), NFP_BPF_ABI_FLAGS); wrp_immed(nfp_prog, reg_b(2), 0x41221211); wrp_immed(nfp_prog, reg_b(3), 0x41001211); @@ -1589,7 +2376,7 @@ static void nfp_outro_tc_da(struct nfp_prog *nfp_prog) 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_br_relo(nfp_prog, BR_UNC, BR_OFF_RELO, 2, RELO_BR_NEXT_PKT); emit_shf(nfp_prog, reg_b(2), reg_a(2), SHF_OP_OR, reg_b(2), SHF_SC_L_SHF, 4); @@ -1608,10 +2395,9 @@ static void nfp_outro_xdp(struct nfp_prog *nfp_prog) /* Target for aborts */ nfp_prog->tgt_abort = nfp_prog_current_offset(nfp_prog); - emit_br_def(nfp_prog, nfp_prog->tgt_done, 2); + emit_br_relo(nfp_prog, BR_UNC, BR_OFF_RELO, 2, RELO_BR_NEXT_PKT); - emit_alu(nfp_prog, reg_a(0), - reg_none(), ALU_OP_NONE, NFP_BPF_ABI_FLAGS); + wrp_mov(nfp_prog, reg_a(0), NFP_BPF_ABI_FLAGS); emit_ld_field(nfp_prog, reg_a(0), 0xc, reg_imm(0x82), SHF_SC_L_SHF, 16); /* Target for normal exits */ @@ -1630,26 +2416,23 @@ static void nfp_outro_xdp(struct nfp_prog *nfp_prog) emit_shf(nfp_prog, reg_b(2), reg_imm(0xff), SHF_OP_AND, reg_b(2), SHF_SC_R_SHF, 0); - emit_br_def(nfp_prog, nfp_prog->tgt_done, 2); + emit_br_relo(nfp_prog, BR_UNC, BR_OFF_RELO, 2, RELO_BR_NEXT_PKT); - emit_alu(nfp_prog, reg_a(0), - reg_none(), ALU_OP_NONE, NFP_BPF_ABI_FLAGS); + wrp_mov(nfp_prog, reg_a(0), NFP_BPF_ABI_FLAGS); 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: + switch (nfp_prog->type) { + case BPF_PROG_TYPE_SCHED_CLS: nfp_outro_tc_da(nfp_prog); break; - case NN_ACT_TC_DROP: - case NN_ACT_TC_REDIR: - nfp_outro_tc_legacy(nfp_prog); - break; - case NN_ACT_XDP: + case BPF_PROG_TYPE_XDP: nfp_outro_xdp(nfp_prog); break; + default: + WARN_ON(1); } } @@ -1684,33 +2467,17 @@ static int nfp_translate(struct nfp_prog *nfp_prog) nfp_prog->n_translated++; } + nfp_prog->last_bpf_off = nfp_prog_current_offset(nfp_prog) - 1; + 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); - } + wrp_nops(nfp_prog, NFP_USTORE_PREFETCH_WINDOW); + if (nfp_prog->error) + return nfp_prog->error; - return 0; + return nfp_fixup_branches(nfp_prog); } /* --- Optimizations --- */ @@ -1737,38 +2504,6 @@ static void nfp_bpf_opt_reg_init(struct nfp_prog *nfp_prog) } } -/* 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; - } - - 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) { @@ -1802,6 +2537,9 @@ static void nfp_bpf_opt_ld_mask(struct nfp_prog *nfp_prog) if (next.src_reg || next.dst_reg) continue; + if (meta2->flags & FLAG_INSN_IS_JUMP_DST) + continue; + meta2->skip = true; } } @@ -1838,86 +2576,405 @@ static void nfp_bpf_opt_ld_shift(struct nfp_prog *nfp_prog) if (next1.imm != 0x20 || next2.imm != 0x20) continue; + if (meta2->flags & FLAG_INSN_IS_JUMP_DST || + meta3->flags & FLAG_INSN_IS_JUMP_DST) + continue; + meta2->skip = true; meta3->skip = true; } } -static int nfp_bpf_optimize(struct nfp_prog *nfp_prog) +/* load/store pair that forms memory copy sould look like the following: + * + * ld_width R, [addr_src + offset_src] + * st_width [addr_dest + offset_dest], R + * + * The destination register of load and source register of store should + * be the same, load and store should also perform at the same width. + * If either of addr_src or addr_dest is stack pointer, we don't do the + * CPP optimization as stack is modelled by registers on NFP. + */ +static bool +curr_pair_is_memcpy(struct nfp_insn_meta *ld_meta, + struct nfp_insn_meta *st_meta) { - int ret; + struct bpf_insn *ld = &ld_meta->insn; + struct bpf_insn *st = &st_meta->insn; - nfp_bpf_opt_reg_init(nfp_prog); + if (!is_mbpf_load(ld_meta) || !is_mbpf_store(st_meta)) + return false; - ret = nfp_bpf_opt_reg_rename(nfp_prog); - if (ret) - return ret; + if (ld_meta->ptr.type != PTR_TO_PACKET) + return false; + + if (st_meta->ptr.type != PTR_TO_PACKET) + return false; + + if (BPF_SIZE(ld->code) != BPF_SIZE(st->code)) + return false; + + if (ld->dst_reg != st->src_reg) + return false; + + /* There is jump to the store insn in this pair. */ + if (st_meta->flags & FLAG_INSN_IS_JUMP_DST) + return false; + + return true; +} + +/* Currently, we only support chaining load/store pairs if: + * + * - Their address base registers are the same. + * - Their address offsets are in the same order. + * - They operate at the same memory width. + * - There is no jump into the middle of them. + */ +static bool +curr_pair_chain_with_previous(struct nfp_insn_meta *ld_meta, + struct nfp_insn_meta *st_meta, + struct bpf_insn *prev_ld, + struct bpf_insn *prev_st) +{ + u8 prev_size, curr_size, prev_ld_base, prev_st_base, prev_ld_dst; + struct bpf_insn *ld = &ld_meta->insn; + struct bpf_insn *st = &st_meta->insn; + s16 prev_ld_off, prev_st_off; + + /* This pair is the start pair. */ + if (!prev_ld) + return true; + + prev_size = BPF_LDST_BYTES(prev_ld); + curr_size = BPF_LDST_BYTES(ld); + prev_ld_base = prev_ld->src_reg; + prev_st_base = prev_st->dst_reg; + prev_ld_dst = prev_ld->dst_reg; + prev_ld_off = prev_ld->off; + prev_st_off = prev_st->off; + + if (ld->dst_reg != prev_ld_dst) + return false; + + if (ld->src_reg != prev_ld_base || st->dst_reg != prev_st_base) + return false; + + if (curr_size != prev_size) + return false; + + /* There is jump to the head of this pair. */ + if (ld_meta->flags & FLAG_INSN_IS_JUMP_DST) + return false; + + /* Both in ascending order. */ + if (prev_ld_off + prev_size == ld->off && + prev_st_off + prev_size == st->off) + return true; + + /* Both in descending order. */ + if (ld->off + curr_size == prev_ld_off && + st->off + curr_size == prev_st_off) + return true; + + return false; +} + +/* Return TRUE if cross memory access happens. Cross memory access means + * store area is overlapping with load area that a later load might load + * the value from previous store, for this case we can't treat the sequence + * as an memory copy. + */ +static bool +cross_mem_access(struct bpf_insn *ld, struct nfp_insn_meta *head_ld_meta, + struct nfp_insn_meta *head_st_meta) +{ + s16 head_ld_off, head_st_off, ld_off; + + /* Different pointer types does not overlap. */ + if (head_ld_meta->ptr.type != head_st_meta->ptr.type) + return false; + + /* load and store are both PTR_TO_PACKET, check ID info. */ + if (head_ld_meta->ptr.id != head_st_meta->ptr.id) + return true; + + /* Canonicalize the offsets. Turn all of them against the original + * base register. + */ + head_ld_off = head_ld_meta->insn.off + head_ld_meta->ptr.off; + head_st_off = head_st_meta->insn.off + head_st_meta->ptr.off; + ld_off = ld->off + head_ld_meta->ptr.off; + + /* Ascending order cross. */ + if (ld_off > head_ld_off && + head_ld_off < head_st_off && ld_off >= head_st_off) + return true; + + /* Descending order cross. */ + if (ld_off < head_ld_off && + head_ld_off > head_st_off && ld_off <= head_st_off) + return true; + + return false; +} + +/* This pass try to identify the following instructoin sequences. + * + * load R, [regA + offA] + * store [regB + offB], R + * load R, [regA + offA + const_imm_A] + * store [regB + offB + const_imm_A], R + * load R, [regA + offA + 2 * const_imm_A] + * store [regB + offB + 2 * const_imm_A], R + * ... + * + * Above sequence is typically generated by compiler when lowering + * memcpy. NFP prefer using CPP instructions to accelerate it. + */ +static void nfp_bpf_opt_ldst_gather(struct nfp_prog *nfp_prog) +{ + struct nfp_insn_meta *head_ld_meta = NULL; + struct nfp_insn_meta *head_st_meta = NULL; + struct nfp_insn_meta *meta1, *meta2; + struct bpf_insn *prev_ld = NULL; + struct bpf_insn *prev_st = NULL; + u8 count = 0; + + nfp_for_each_insn_walk2(nfp_prog, meta1, meta2) { + struct bpf_insn *ld = &meta1->insn; + struct bpf_insn *st = &meta2->insn; + + /* Reset record status if any of the following if true: + * - The current insn pair is not load/store. + * - The load/store pair doesn't chain with previous one. + * - The chained load/store pair crossed with previous pair. + * - The chained load/store pair has a total size of memory + * copy beyond 128 bytes which is the maximum length a + * single NFP CPP command can transfer. + */ + if (!curr_pair_is_memcpy(meta1, meta2) || + !curr_pair_chain_with_previous(meta1, meta2, prev_ld, + prev_st) || + (head_ld_meta && (cross_mem_access(ld, head_ld_meta, + head_st_meta) || + head_ld_meta->ldst_gather_len >= 128))) { + if (!count) + continue; + + if (count > 1) { + s16 prev_ld_off = prev_ld->off; + s16 prev_st_off = prev_st->off; + s16 head_ld_off = head_ld_meta->insn.off; + + if (prev_ld_off < head_ld_off) { + head_ld_meta->insn.off = prev_ld_off; + head_st_meta->insn.off = prev_st_off; + head_ld_meta->ldst_gather_len = + -head_ld_meta->ldst_gather_len; + } + + head_ld_meta->paired_st = &head_st_meta->insn; + head_st_meta->skip = true; + } else { + head_ld_meta->ldst_gather_len = 0; + } + + /* If the chain is ended by an load/store pair then this + * could serve as the new head of the the next chain. + */ + if (curr_pair_is_memcpy(meta1, meta2)) { + head_ld_meta = meta1; + head_st_meta = meta2; + head_ld_meta->ldst_gather_len = + BPF_LDST_BYTES(ld); + meta1 = nfp_meta_next(meta1); + meta2 = nfp_meta_next(meta2); + prev_ld = ld; + prev_st = st; + count = 1; + } else { + head_ld_meta = NULL; + head_st_meta = NULL; + prev_ld = NULL; + prev_st = NULL; + count = 0; + } + + continue; + } + + if (!head_ld_meta) { + head_ld_meta = meta1; + head_st_meta = meta2; + } else { + meta1->skip = true; + meta2->skip = true; + } + + head_ld_meta->ldst_gather_len += BPF_LDST_BYTES(ld); + meta1 = nfp_meta_next(meta1); + meta2 = nfp_meta_next(meta2); + prev_ld = ld; + prev_st = st; + count++; + } +} + +static int nfp_bpf_optimize(struct nfp_prog *nfp_prog) +{ + nfp_bpf_opt_reg_init(nfp_prog); nfp_bpf_opt_ld_mask(nfp_prog); nfp_bpf_opt_ld_shift(nfp_prog); + nfp_bpf_opt_ldst_gather(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) +static int nfp_bpf_ustore_calc(u64 *prog, unsigned int len) { - struct nfp_prog *nfp_prog; - int ret; + __le64 *ustore = (__force __le64 *)prog; + int i; - nfp_prog = kzalloc(sizeof(*nfp_prog), GFP_KERNEL); - if (!nfp_prog) - return -ENOMEM; + for (i = 0; i < len; i++) { + int err; - INIT_LIST_HEAD(&nfp_prog->insns); - nfp_prog->act = act; - nfp_prog->start_off = prog_start; - nfp_prog->tgt_done = prog_done; + err = nfp_ustore_check_valid_no_ecc(prog[i]); + if (err) + return err; - ret = nfp_prog_prepare(nfp_prog, filter->insnsi, filter->len); - if (ret) - goto out; + ustore[i] = cpu_to_le64(nfp_ustore_calc_ecc_insn(prog[i])); + } - ret = nfp_prog_verify(nfp_prog, filter); - if (ret) - goto out; + return 0; +} - ret = nfp_bpf_optimize(nfp_prog); - if (ret) - goto out; +static void nfp_bpf_prog_trim(struct nfp_prog *nfp_prog) +{ + void *prog; - if (nfp_prog->num_regs <= 7) - nfp_prog->regs_per_thread = 16; - else - nfp_prog->regs_per_thread = 32; + prog = kvmalloc_array(nfp_prog->prog_len, sizeof(u64), GFP_KERNEL); + if (!prog) + return; + + nfp_prog->__prog_alloc_len = nfp_prog->prog_len * sizeof(u64); + memcpy(prog, nfp_prog->prog, nfp_prog->__prog_alloc_len); + kvfree(nfp_prog->prog); + nfp_prog->prog = prog; +} + +int nfp_bpf_jit(struct nfp_prog *nfp_prog) +{ + int ret; - nfp_prog->prog = prog_mem; - nfp_prog->__prog_alloc_len = prog_sz; + ret = nfp_bpf_optimize(nfp_prog); + if (ret) + return ret; ret = nfp_translate(nfp_prog); if (ret) { pr_err("Translation failed with error %d (translated: %u)\n", ret, nfp_prog->n_translated); - ret = -EINVAL; + return -EINVAL; } - res->n_instr = nfp_prog->prog_len; - res->dense_mode = nfp_prog->num_regs <= 7; -out: - nfp_prog_free(nfp_prog); + nfp_bpf_prog_trim(nfp_prog); return ret; } + +void nfp_bpf_jit_prepare(struct nfp_prog *nfp_prog, unsigned int cnt) +{ + struct nfp_insn_meta *meta; + + /* Another pass to record jump information. */ + list_for_each_entry(meta, &nfp_prog->insns, l) { + u64 code = meta->insn.code; + + if (BPF_CLASS(code) == BPF_JMP && BPF_OP(code) != BPF_EXIT && + BPF_OP(code) != BPF_CALL) { + struct nfp_insn_meta *dst_meta; + unsigned short dst_indx; + + dst_indx = meta->n + 1 + meta->insn.off; + dst_meta = nfp_bpf_goto_meta(nfp_prog, meta, dst_indx, + cnt); + + meta->jmp_dst = dst_meta; + dst_meta->flags |= FLAG_INSN_IS_JUMP_DST; + } + } +} + +bool nfp_bpf_supported_opcode(u8 code) +{ + return !!instr_cb[code]; +} + +void *nfp_bpf_relo_for_vnic(struct nfp_prog *nfp_prog, struct nfp_bpf_vnic *bv) +{ + unsigned int i; + u64 *prog; + int err; + + prog = kmemdup(nfp_prog->prog, nfp_prog->prog_len * sizeof(u64), + GFP_KERNEL); + if (!prog) + return ERR_PTR(-ENOMEM); + + for (i = 0; i < nfp_prog->prog_len; i++) { + enum nfp_relo_type special; + u32 val; + + special = FIELD_GET(OP_RELO_TYPE, prog[i]); + switch (special) { + case RELO_NONE: + continue; + case RELO_BR_REL: + br_add_offset(&prog[i], bv->start_off); + break; + case RELO_BR_GO_OUT: + br_set_offset(&prog[i], + nfp_prog->tgt_out + bv->start_off); + break; + case RELO_BR_GO_ABORT: + br_set_offset(&prog[i], + nfp_prog->tgt_abort + bv->start_off); + break; + case RELO_BR_NEXT_PKT: + br_set_offset(&prog[i], bv->tgt_done); + break; + case RELO_BR_HELPER: + val = br_get_offset(prog[i]); + val -= BR_OFF_RELO; + switch (val) { + case BPF_FUNC_map_lookup_elem: + val = nfp_prog->bpf->helpers.map_lookup; + break; + default: + pr_err("relocation of unknown helper %d\n", + val); + err = -EINVAL; + goto err_free_prog; + } + br_set_offset(&prog[i], val); + break; + case RELO_IMMED_REL: + immed_add_value(&prog[i], bv->start_off); + break; + } + + prog[i] &= ~OP_RELO_TYPE; + } + + err = nfp_bpf_ustore_calc(prog, nfp_prog->prog_len); + if (err) + goto err_free_prog; + + return prog; + +err_free_prog: + kfree(prog); + return ERR_PTR(err); +} |