OSDN Git Service

cxgb4: add support to create hash-filters via tc-flower offload
authorKumar Sanghvi <kumaras@chelsio.com>
Wed, 1 Nov 2017 03:23:05 +0000 (08:53 +0530)
committerDavid S. Miller <davem@davemloft.net>
Wed, 1 Nov 2017 13:06:03 +0000 (22:06 +0900)
Determine whether the flow classifies as exact-match with respect to
4-tuple and configured tuple mask in hw. If successfully classified
as exact-match, offload the flow as hash-filter in hw.

Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.h
drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c

index 9b3ff62..abab67d 100644 (file)
@@ -31,6 +31,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
+#include <net/ipv6.h>
 
 #include "cxgb4.h"
 #include "t4_regs.h"
@@ -765,6 +766,153 @@ static void fill_default_mask(struct ch_filter_specification *fs)
                fs->mask.fport = ~0;
 }
 
+static bool is_addr_all_mask(u8 *ipmask, int family)
+{
+       if (family == AF_INET) {
+               struct in_addr *addr;
+
+               addr = (struct in_addr *)ipmask;
+               if (addr->s_addr == 0xffffffff)
+                       return true;
+       } else if (family == AF_INET6) {
+               struct in6_addr *addr6;
+
+               addr6 = (struct in6_addr *)ipmask;
+               if (addr6->s6_addr32[0] == 0xffffffff &&
+                   addr6->s6_addr32[1] == 0xffffffff &&
+                   addr6->s6_addr32[2] == 0xffffffff &&
+                   addr6->s6_addr32[3] == 0xffffffff)
+                       return true;
+       }
+       return false;
+}
+
+static bool is_inaddr_any(u8 *ip, int family)
+{
+       int addr_type;
+
+       if (family == AF_INET) {
+               struct in_addr *addr;
+
+               addr = (struct in_addr *)ip;
+               if (addr->s_addr == htonl(INADDR_ANY))
+                       return true;
+       } else if (family == AF_INET6) {
+               struct in6_addr *addr6;
+
+               addr6 = (struct in6_addr *)ip;
+               addr_type = ipv6_addr_type((const struct in6_addr *)
+                                          &addr6);
+               if (addr_type == IPV6_ADDR_ANY)
+                       return true;
+       }
+       return false;
+}
+
+bool is_filter_exact_match(struct adapter *adap,
+                          struct ch_filter_specification *fs)
+{
+       struct tp_params *tp = &adap->params.tp;
+       u64 hash_filter_mask = tp->hash_filter_mask;
+       u32 mask;
+
+       if (!is_hashfilter(adap))
+               return false;
+
+       if (fs->type) {
+               if (is_inaddr_any(fs->val.fip, AF_INET6) ||
+                   !is_addr_all_mask(fs->mask.fip, AF_INET6))
+                       return false;
+
+               if (is_inaddr_any(fs->val.lip, AF_INET6) ||
+                   !is_addr_all_mask(fs->mask.lip, AF_INET6))
+                       return false;
+       } else {
+               if (is_inaddr_any(fs->val.fip, AF_INET) ||
+                   !is_addr_all_mask(fs->mask.fip, AF_INET))
+                       return false;
+
+               if (is_inaddr_any(fs->val.lip, AF_INET) ||
+                   !is_addr_all_mask(fs->mask.lip, AF_INET))
+                       return false;
+       }
+
+       if (!fs->val.lport || fs->mask.lport != 0xffff)
+               return false;
+
+       if (!fs->val.fport || fs->mask.fport != 0xffff)
+               return false;
+
+       if (tp->fcoe_shift >= 0) {
+               mask = (hash_filter_mask >> tp->fcoe_shift) & FT_FCOE_W;
+               if (mask && !fs->mask.fcoe)
+                       return false;
+       }
+
+       if (tp->port_shift >= 0) {
+               mask = (hash_filter_mask >> tp->port_shift) & FT_PORT_W;
+               if (mask && !fs->mask.iport)
+                       return false;
+       }
+
+       if (tp->vnic_shift >= 0) {
+               mask = (hash_filter_mask >> tp->vnic_shift) & FT_VNIC_ID_W;
+
+               if ((adap->params.tp.ingress_config & VNIC_F)) {
+                       if (mask && !fs->mask.pfvf_vld)
+                               return false;
+               } else {
+                       if (mask && !fs->mask.ovlan_vld)
+                               return false;
+               }
+       }
+
+       if (tp->vlan_shift >= 0) {
+               mask = (hash_filter_mask >> tp->vlan_shift) & FT_VLAN_W;
+               if (mask && !fs->mask.ivlan)
+                       return false;
+       }
+
+       if (tp->tos_shift >= 0) {
+               mask = (hash_filter_mask >> tp->tos_shift) & FT_TOS_W;
+               if (mask && !fs->mask.tos)
+                       return false;
+       }
+
+       if (tp->protocol_shift >= 0) {
+               mask = (hash_filter_mask >> tp->protocol_shift) & FT_PROTOCOL_W;
+               if (mask && !fs->mask.proto)
+                       return false;
+       }
+
+       if (tp->ethertype_shift >= 0) {
+               mask = (hash_filter_mask >> tp->ethertype_shift) &
+                       FT_ETHERTYPE_W;
+               if (mask && !fs->mask.ethtype)
+                       return false;
+       }
+
+       if (tp->macmatch_shift >= 0) {
+               mask = (hash_filter_mask >> tp->macmatch_shift) & FT_MACMATCH_W;
+               if (mask && !fs->mask.macidx)
+                       return false;
+       }
+
+       if (tp->matchtype_shift >= 0) {
+               mask = (hash_filter_mask >> tp->matchtype_shift) &
+                       FT_MPSHITTYPE_W;
+               if (mask && !fs->mask.matchtype)
+                       return false;
+       }
+       if (tp->frag_shift >= 0) {
+               mask = (hash_filter_mask >> tp->frag_shift) &
+                       FT_FRAGMENTATION_W;
+               if (mask && !fs->mask.frag)
+                       return false;
+       }
+       return true;
+}
+
 static u64 hash_filter_ntuple(struct ch_filter_specification *fs,
                              struct net_device *dev)
 {
index 9475abd..8db5fca 100644 (file)
@@ -51,4 +51,6 @@ int delete_filter(struct adapter *adapter, unsigned int fidx);
 int writable_filter(struct filter_entry *f);
 void clear_all_filters(struct adapter *adapter);
 int init_hash_filter(struct adapter *adap);
+bool is_filter_exact_match(struct adapter *adap,
+                          struct ch_filter_specification *fs);
 #endif /* __CXGB4_FILTER_H */
index 3953bc1..d4a548a 100644 (file)
@@ -38,6 +38,7 @@
 #include <net/tc_act/tc_vlan.h>
 
 #include "cxgb4.h"
+#include "cxgb4_filter.h"
 #include "cxgb4_tc_flower.h"
 
 #define STATS_CHECK_PERIOD (HZ / 2)
@@ -672,11 +673,16 @@ int cxgb4_tc_flower_replace(struct net_device *dev,
        cxgb4_process_flow_match(dev, cls, fs);
        cxgb4_process_flow_actions(dev, cls, fs);
 
-       fidx = cxgb4_get_free_ftid(dev, fs->type ? PF_INET6 : PF_INET);
-       if (fidx < 0) {
-               netdev_err(dev, "%s: No fidx for offload.\n", __func__);
-               ret = -ENOMEM;
-               goto free_entry;
+       fs->hash = is_filter_exact_match(adap, fs);
+       if (fs->hash) {
+               fidx = 0;
+       } else {
+               fidx = cxgb4_get_free_ftid(dev, fs->type ? PF_INET6 : PF_INET);
+               if (fidx < 0) {
+                       netdev_err(dev, "%s: No fidx for offload.\n", __func__);
+                       ret = -ENOMEM;
+                       goto free_entry;
+               }
        }
 
        init_completion(&ctx.completion);