| Message ID | 20241111104327.65539-1-dceara@redhat.com |
|---|---|
| State | New |
| Headers | show |
| Series | [ovs-dev] logical-fields: Reuse registers for ct_*_dst() actions. | expand |
| Context | Check | Description |
|---|---|---|
| ovsrobot/apply-robot | success | apply and check: success |
| ovsrobot/github-robot-_Build_and_Test | success | github build: passed |
| ovsrobot/github-robot-_ovn-kubernetes | fail | github build: failed |
| ovsrobot/github-robot-_Build_and_Test | success | github build: passed |
| ovsrobot/github-robot-_ovn-kubernetes | success | github build: passed |
Recheck-request: github-robot On 11/11/24 11:43 AM, Dumitru Ceara wrote: > The actions that retrieve the original tuple destination IPs and ports > were used in the following way in ovn-northd: > > REG_ORIG_DIP_IPV4 " = ct_nw_dst() > REG_ORIG_DIP_IPV6 " = ct_ip6_dst() > REG_ORIG_TP_DPORT " = ct_tp_dst() > > Where: > REG_ORIG_DIP_IPV4 is reg1 > REG_ORIG_DIP_IPV6 is xxreg1 > REG_ORIG_TP_DPORT is reg2[0..15] > > While the actions use a different intermediate register: > ct_nw_dst() was reg4<-0; reg4<-ct_nw_dst > ct_ip6_dst() was xxreg0<-0; xxreg0<-ct_ip6_dst > ct_tp_dst() was reg8[0..15]<-0; reg8[0..15]<-ct_tp_dst > > We can reduce the set of registers we use in the OVN pipeline by > changing the action definition to use the same registers ovn-northd uses > as final destination for the ct_*_dst() values. This will generate the > following openflow rules: > > REG_ORIG_DIP_IPV4 " = ct_nw_dst() > reg1<-0; reg1<-ct_nw_dst; reg1<-reg1 > REG_ORIG_DIP_IPV6 " = ct_ip6_dst() > xxreg1<-0; xxreg1<-ct_ip6_dst; xxreg1<-xxreg1 > REG_ORIG_TP_DPORT " = ct_tp_dst() > reg2[0..15]<-0; reg2[0..15]<-ct_tp_dst; reg2[0..15]<-reg2[0..15] > > As Ilya Maximets points out, overlapping source and destination are > well defined for move actions: > https://opennetworking.org/wp-content/uploads/2014/10/openflow-switch-v1.5.0.noipr.pdf > > This action must behaves properly when src_oxm_id overlaps with > dst_oxm_id, that is, it behaves as if src_oxm_id were copied out > to a temporary buffer, then the temporary buffer copied to > dst_oxm_id, if this is not possible the switch must reject the > Copy Field action with a Bad Set Type error message. > > OpenvSwitch doesn't reject such actions. > > Fixes: 0f806cf08c36 ("Fix load balanced hairpin traffic for fragmented packets.") > Signed-off-by: Dumitru Ceara <dceara@redhat.com> > ---
diff --git a/include/ovn/logical-fields.h b/include/ovn/logical-fields.h index d563e044cb..70c6b93c41 100644 --- a/include/ovn/logical-fields.h +++ b/include/ovn/logical-fields.h @@ -60,9 +60,10 @@ enum ovn_controller_event { #define MFF_LOG_LB_AFF_MATCH_LR_IP6_ADDR MFF_XXREG1 #define MFF_LOG_LB_AFF_MATCH_PORT MFF_REG8 -#define MFF_LOG_CT_ORIG_NW_DST_ADDR MFF_REG4 -#define MFF_LOG_CT_ORIG_IP6_DST_ADDR MFF_XXREG0 -#define MFF_LOG_CT_ORIG_TP_DST_PORT MFF_REG8 +#define MFF_LOG_CT_ORIG_NW_DST_ADDR MFF_REG1 /* REG_ORIG_DIP_IPV4 */ +#define MFF_LOG_CT_ORIG_IP6_DST_ADDR MFF_XXREG1 /* REG_ORIG_DIP_IPV6 */ +#define MFF_LOG_CT_ORIG_TP_DST_PORT MFF_REG2 /* REG_ORIG_TP_DPORT + * (bits 0..15). */ void ovn_init_symtab(struct shash *symtab); diff --git a/tests/ovn.at b/tests/ovn.at index 15b78f4c37..03fd2090fc 100644 --- a/tests/ovn.at +++ b/tests/ovn.at @@ -2420,10 +2420,10 @@ mac_cache_use; # ct_nw_dst() reg1 = ct_nw_dst(); - encodes as set_field:0->reg4,resubmit(,OFTABLE_CT_ORIG_NW_DST_LOAD),move:NXM_NX_REG4[[]]->NXM_NX_XXREG0[[64..95]] + encodes as set_field:0->reg1,resubmit(,OFTABLE_CT_ORIG_NW_DST_LOAD),move:NXM_NX_REG1[[]]->NXM_NX_XXREG0[[64..95]] xreg1[[3..34]] = ct_nw_dst(); - encodes as set_field:0->reg4,resubmit(,OFTABLE_CT_ORIG_NW_DST_LOAD),move:NXM_NX_REG4[[]]->NXM_NX_XXREG0[[3..34]] + encodes as set_field:0->reg1,resubmit(,OFTABLE_CT_ORIG_NW_DST_LOAD),move:NXM_NX_REG1[[]]->NXM_NX_XXREG0[[3..34]] reg1[[3..34]] = ct_nw_dst(); Cannot select bits 3 to 34 of 32-bit field reg1. @@ -2442,7 +2442,7 @@ ct_nw_dst(); # ct_ip6_dst() xxreg1 = ct_ip6_dst(); - encodes as set_field:0/0xffffffffffffffff->xxreg0,set_field:0/0xffffffffffffffff0000000000000000->xxreg0,resubmit(,OFTABLE_CT_ORIG_IP6_DST_LOAD),move:NXM_NX_XXREG0[[]]->NXM_NX_XXREG1[[]] + encodes as set_field:0/0xffffffffffffffff->xxreg1,set_field:0/0xffffffffffffffff0000000000000000->xxreg1,resubmit(,OFTABLE_CT_ORIG_IP6_DST_LOAD),move:NXM_NX_XXREG1[[]]->NXM_NX_XXREG1[[]] reg1 = ct_ip6_dst(); Cannot use 32-bit field reg1[[0..31]] where 128-bit field is required. @@ -2455,7 +2455,7 @@ ct_ip6_dst(); # ct_tp_dst() reg1[[0..15]] = ct_tp_dst(); - encodes as set_field:0/0xffff->reg8,resubmit(,OFTABLE_CT_ORIG_TP_DST_LOAD),move:NXM_NX_REG8[[0..15]]->NXM_NX_XXREG0[[64..79]] + encodes as set_field:0/0xffff->reg2,resubmit(,OFTABLE_CT_ORIG_TP_DST_LOAD),move:NXM_NX_REG2[[0..15]]->NXM_NX_XXREG0[[64..79]] reg1 = ct_tp_dst(); Cannot use 32-bit field reg1[[0..31]] where 16-bit field is required.
The actions that retrieve the original tuple destination IPs and ports were used in the following way in ovn-northd: REG_ORIG_DIP_IPV4 " = ct_nw_dst() REG_ORIG_DIP_IPV6 " = ct_ip6_dst() REG_ORIG_TP_DPORT " = ct_tp_dst() Where: REG_ORIG_DIP_IPV4 is reg1 REG_ORIG_DIP_IPV6 is xxreg1 REG_ORIG_TP_DPORT is reg2[0..15] While the actions use a different intermediate register: ct_nw_dst() was reg4<-0; reg4<-ct_nw_dst ct_ip6_dst() was xxreg0<-0; xxreg0<-ct_ip6_dst ct_tp_dst() was reg8[0..15]<-0; reg8[0..15]<-ct_tp_dst We can reduce the set of registers we use in the OVN pipeline by changing the action definition to use the same registers ovn-northd uses as final destination for the ct_*_dst() values. This will generate the following openflow rules: REG_ORIG_DIP_IPV4 " = ct_nw_dst() reg1<-0; reg1<-ct_nw_dst; reg1<-reg1 REG_ORIG_DIP_IPV6 " = ct_ip6_dst() xxreg1<-0; xxreg1<-ct_ip6_dst; xxreg1<-xxreg1 REG_ORIG_TP_DPORT " = ct_tp_dst() reg2[0..15]<-0; reg2[0..15]<-ct_tp_dst; reg2[0..15]<-reg2[0..15] As Ilya Maximets points out, overlapping source and destination are well defined for move actions: https://opennetworking.org/wp-content/uploads/2014/10/openflow-switch-v1.5.0.noipr.pdf This action must behaves properly when src_oxm_id overlaps with dst_oxm_id, that is, it behaves as if src_oxm_id were copied out to a temporary buffer, then the temporary buffer copied to dst_oxm_id, if this is not possible the switch must reject the Copy Field action with a Bad Set Type error message. OpenvSwitch doesn't reject such actions. Fixes: 0f806cf08c36 ("Fix load balanced hairpin traffic for fragmented packets.") Signed-off-by: Dumitru Ceara <dceara@redhat.com> --- include/ovn/logical-fields.h | 7 ++++--- tests/ovn.at | 8 ++++---- 2 files changed, 8 insertions(+), 7 deletions(-)