| Message ID | 20191031084749.14626-3-bjorn.topel@gmail.com |
|---|---|
| State | Changes Requested |
| Delegated to: | BPF Maintainers |
| Headers | show |
| Series | xsk: XSKMAP performance improvements | expand |
On Thu, Oct 31, 2019 at 09:47:48AM +0100, Björn Töpel wrote: > From: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > > Inline the xsk_map_lookup_elem() via implementing the map_gen_lookup() > callback. This results in emitting the bpf instructions in place of > bpf_map_lookup_elem() helper call and better performance of bpf > programs. > > Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com> > Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > --- > kernel/bpf/xskmap.c | 17 +++++++++++++++++ > 1 file changed, 17 insertions(+) > > diff --git a/kernel/bpf/xskmap.c b/kernel/bpf/xskmap.c > index edcbd863650e..fa32f775b4de 100644 > --- a/kernel/bpf/xskmap.c > +++ b/kernel/bpf/xskmap.c > @@ -163,6 +163,22 @@ struct xdp_sock *__xsk_map_lookup_elem(struct bpf_map *map, u32 key) > return xs; > } > > +static u32 xsk_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf) > +{ > + const int ret = BPF_REG_0, mp = BPF_REG_1, index = BPF_REG_2; > + struct bpf_insn *insn = insn_buf; > + > + *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0); > + *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5); > + *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(sizeof(struct xsk_sock *))); > + *insn++ = BPF_ALU64_IMM(BPF_ADD, mp, offsetof(struct xsk_map, xsk_map)); > + *insn++ = BPF_ALU64_REG(BPF_ADD, ret, mp); > + *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0); Your map slots are always exactly sizeof(struct xdp_sock *), right? Wouldn't this BPF_DW crash on 32 bit? Meaning, it would have to be BPF_LDX_MEM(BPF_SIZEOF(struct xsk_sock *), ...)? > + *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1); > + *insn++ = BPF_MOV64_IMM(ret, 0); > + return insn - insn_buf; > +} > + > int __xsk_map_redirect(struct bpf_map *map, struct xdp_buff *xdp, > struct xdp_sock *xs) > { > @@ -303,6 +319,7 @@ const struct bpf_map_ops xsk_map_ops = { > .map_free = xsk_map_free, > .map_get_next_key = xsk_map_get_next_key, > .map_lookup_elem = xsk_map_lookup_elem, > + .map_gen_lookup = xsk_map_gen_lookup, > .map_lookup_elem_sys_only = xsk_map_lookup_elem_sys_only, > .map_update_elem = xsk_map_update_elem, > .map_delete_elem = xsk_map_delete_elem, > -- > 2.20.1 >
On Fri, 1 Nov 2019 at 00:48, Daniel Borkmann <daniel@iogearbox.net> wrote: > > On Thu, Oct 31, 2019 at 09:47:48AM +0100, Björn Töpel wrote: > > From: Maciej Fijalkowski <maciej.fijalkowski@intel.com> [...] > > +static u32 xsk_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf) > > +{ > > + const int ret = BPF_REG_0, mp = BPF_REG_1, index = BPF_REG_2; > > + struct bpf_insn *insn = insn_buf; > > + > > + *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0); > > + *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5); > > + *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(sizeof(struct xsk_sock *))); > > + *insn++ = BPF_ALU64_IMM(BPF_ADD, mp, offsetof(struct xsk_map, xsk_map)); > > + *insn++ = BPF_ALU64_REG(BPF_ADD, ret, mp); > > + *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0); > > Your map slots are always exactly sizeof(struct xdp_sock *), right? Wouldn't > this BPF_DW crash on 32 bit? > > Meaning, it would have to be BPF_LDX_MEM(BPF_SIZEOF(struct xsk_sock *), ...)? > Indeed. Thanks for finding this. I'll do a respin. Björn
diff --git a/kernel/bpf/xskmap.c b/kernel/bpf/xskmap.c index edcbd863650e..fa32f775b4de 100644 --- a/kernel/bpf/xskmap.c +++ b/kernel/bpf/xskmap.c @@ -163,6 +163,22 @@ struct xdp_sock *__xsk_map_lookup_elem(struct bpf_map *map, u32 key) return xs; } +static u32 xsk_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf) +{ + const int ret = BPF_REG_0, mp = BPF_REG_1, index = BPF_REG_2; + struct bpf_insn *insn = insn_buf; + + *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0); + *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5); + *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(sizeof(struct xsk_sock *))); + *insn++ = BPF_ALU64_IMM(BPF_ADD, mp, offsetof(struct xsk_map, xsk_map)); + *insn++ = BPF_ALU64_REG(BPF_ADD, ret, mp); + *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0); + *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1); + *insn++ = BPF_MOV64_IMM(ret, 0); + return insn - insn_buf; +} + int __xsk_map_redirect(struct bpf_map *map, struct xdp_buff *xdp, struct xdp_sock *xs) { @@ -303,6 +319,7 @@ const struct bpf_map_ops xsk_map_ops = { .map_free = xsk_map_free, .map_get_next_key = xsk_map_get_next_key, .map_lookup_elem = xsk_map_lookup_elem, + .map_gen_lookup = xsk_map_gen_lookup, .map_lookup_elem_sys_only = xsk_map_lookup_elem_sys_only, .map_update_elem = xsk_map_update_elem, .map_delete_elem = xsk_map_delete_elem,