Revert "RDMA/rxe: Remove VLAN code leftovers from RXE"
Subsystems (2)
PATCH v2 1/1 Revert "RDMA/rxe: Remove VLAN code leftovers from RXE"
From: Martin Wilck <[email protected]>
This reverts commit b2d2440430c0fdd5e0cad3efd6d1c9e3d3d02e5b.
It's true that creating rxe on top of 802.1q interfaces doesn't work. Thus, commit fd49ddaf7e26 ("RDMA/rxe: prevent rxe creation on top of vlan interface") was absolutely correct.
But b2d2440430c0 was incorrect assuming that with this change, RDMA and VLAN don't work togehter at all. It just has to be set up differently. Rather than creating rxe on top of the VLAN interface, rxe must be created on top of the physical interface. RDMA then works just fine through VLAN interfaces on top of that physical interface, via the "upper device" logic.
I've tested this mainly with NVMe over RDMA and rping, but I don't see why it wouldn't work just as well for other protocols. If there are real issues, I'd like to know.
b2d2440430c0 broke this setup deliberately and should thus be reverted. Also, b2d2440430c0 removed rxe_dma_device() (which is indeed not necessary), but not its declaration in rxe_loc.h.
--- drivers/infiniband/sw/rxe/rxe_loc.h | 1 - drivers/infiniband/sw/rxe/rxe_net.c | 6 ++++++ drivers/infiniband/sw/rxe/rxe_resp.c | 5 +++++ 3 files changed, 11 insertions(+), 1 deletion(-)
drivers/infiniband/sw/rxe/rxe_loc.h +0 -1
@@ -116,7 +116,6 @@ struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av, int paylen, struct rxe_pkt_info *pkt);int rxe_prepare (struct rxe_pkt_info *pkt,struct sk_buff *skb,u32 *crc);const char *rxe_parent_name (struct rxe_dev *rxe,unsigned int port_num); -struct device *rxe_dma_device (struct rxe_dev *rxe);int rxe_mcast_add (struct rxe_dev *rxe,union ib_gid *mgid);int rxe_mcast_delete (struct rxe_dev *rxe,union ib_gid *mgid);
drivers/infiniband/sw/rxe/rxe_net.c +6 -0
@@ -8,6 +8,7 @@#include <linux/if_arp.h> #include <linux/netdevice.h> #include <linux/if.h> +#include <linux/if_vlan.h> #include <net/udp_tunnel.h> #include <net/sch_generic.h> #include <linux/netfilter.h>
@@ -153,9 +154,14 @@ static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb) {struct udphdr *udph;struct net_device *ndev = skb->dev; +struct net_device *rdev = ndev;struct rxe_dev *rxe =rxe_get_dev_from_net (ndev);struct rxe_pkt_info *pkt =SKB_TO_PKT (skb); +if (!rxe &&is_vlan_dev (rdev)) { + rdev =vlan_dev_real_dev (ndev); + rxe =rxe_get_dev_from_net (rdev); + }if (!rxe) goto drop;
drivers/infiniband/sw/rxe/rxe_resp.c +5 -0
@@ -872,6 +872,11 @@ static enum resp_states do_complete(struct rxe_qp *qp,else wc->network_hdr_type =RDMA_NETWORK_IPV6 ; +if (is_vlan_dev (skb->dev)) { + wc->wc_flags |=IB_WC_WITH_VLAN ; + wc->vlan_id =vlan_dev_vlan_id (skb->dev); + } +if (pkt->mask &RXE_IMMDT_MASK ) { wc->wc_flags |=IB_WC_WITH_IMM ; wc->ex.imm_data =immdt_imm (pkt);
Signature · 1 line
2.29.2
Quoted 29 lines · On Wed, Jan 20, 2021 at 05:19:13PM +0100, [email protected] wrote:
From: Martin Wilck <[email protected]>
This reverts commit
b2d2440430c0fdd5e0cad3efd6d1c9e3d3d02e5b.It's true that creating rxe on top of 802.1q interfaces doesn't work. Thus, commit
fd49ddaf7e26("RDMA/rxe: prevent rxe creation on top of vlan interface") was absolutely correct.But
b2d2440430c0was incorrect assuming that with this change, RDMA and VLAN don't work togehter at all. It just has to be set up differently. Rather than creating rxe on top of the VLAN interface, rxe must be created on top of the physical interface. RDMA then works just fine through VLAN interfaces on top of that physical interface, via the "upper device" logic.I've tested this mainly with NVMe over RDMA and rping, but I don't see why it wouldn't work just as well for other protocols. If there are real issues, I'd like to know.
b2d2440430c0broke this setup deliberately and should thus be reverted. Also,b2d2440430c0removedrxe_dma_device()(which is indeed not necessary), but not its declaration inrxe_loc.h.Fixes b2d2440430c0 ("RDMA/rxe: Remove VLAN code leftovers from RXE")Cc Zhu Yanjun <[email protected]>Cc Mohammad Heib <[email protected]>Cc Nicolas Morey-Chaisemartin <[email protected]>Signed-off-by Martin Wilck <[email protected]>
Quoted 5 lines
--- drivers/infiniband/sw/rxe/rxe_loc.h | 1 - drivers/infiniband/sw/rxe/rxe_net.c | 6 ++++++ drivers/infiniband/sw/rxe/rxe_resp.c | 5 +++++ 3 files changed, 11 insertions(+), 1 deletion(-)
Applied to for-rc, thanks
Jason
Jason Gunthorpe
maintainerQuoted 38 lines · On Wed, Jan 20, 2021 at 05:19:13PM +0100, [email protected] wrote:
From: Martin Wilck <[email protected]>
This reverts commit
b2d2440430c0fdd5e0cad3efd6d1c9e3d3d02e5b.It's true that creating rxe on top of 802.1q interfaces doesn't work. Thus, commit
fd49ddaf7e26("RDMA/rxe: prevent rxe creation on top of vlan interface") was absolutely correct.But
b2d2440430c0was incorrect assuming that with this change, RDMA and VLAN don't work togehter at all. It just has to be set up differently. Rather than creating rxe on top of the VLAN interface, rxe must be created on top of the physical interface. RDMA then works just fine through VLAN interfaces on top of that physical interface, via the "upper device" logic.I've tested this mainly with NVMe over RDMA and rping, but I don't see why it wouldn't work just as well for other protocols. If there are real issues, I'd like to know.
b2d2440430c0broke this setup deliberately and should thus be reverted. Also,b2d2440430c0removedrxe_dma_device()(which is indeed not necessary), but not its declaration inrxe_loc.h.Fixes b2d2440430c0 ("RDMA/rxe: Remove VLAN code leftovers from RXE")Cc Zhu Yanjun <[email protected]>Cc Mohammad Heib <[email protected]>Cc Nicolas Morey-Chaisemartin <[email protected]>Signed-off-by Martin Wilck <[email protected]>drivers/infiniband/sw/rxe/rxe_loc.h | 1 - drivers/infiniband/sw/rxe/rxe_net.c | 6 ++++++ drivers/infiniband/sw/rxe/rxe_resp.c | 5 +++++ 3 files changed, 11 insertions(+), 1 deletion(-)drivers/infiniband/sw/rxe/rxe_loc.h +1
-0+++ b/drivers/infiniband/sw/rxe/rxe_loc.h
drivers/infiniband/sw/rxe/rxe_loc.h struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av, · line 120
Quoted 6 lines
drivers/infiniband/sw/rxe/rxe_loc.h +0
-1@@ -116,7 +116,6 @@ struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av, int paylen, struct rxe_pkt_info *pkt);int rxe_prepare (struct rxe_pkt_info *pkt,struct sk_buff *skb,u32 *crc);const char *rxe_parent_name (struct rxe_dev *rxe,unsigned int port_num); -struct device *rxe_dma_device (struct rxe_dev *rxe);int rxe_mcast_add (struct rxe_dev *rxe,union ib_gid *mgid);int rxe_mcast_delete (struct rxe_dev *rxe,union ib_gid *mgid);
I can't take this hunk as a -rc patch
But I took the rest to for-rc, thanks
Jason