linux-next: manual merge of the rdma tree with the rdma-fixes tree

DISCUSSION Stephen Rothwell 2 participants 4 messages Mar 6, 2025 1 message not in the archive

Subsystems (3)

Linux-next tree InfiniBand Soft-RoCE driver (rxe)

lore

Share & tools

lore

Stephen Rothwell

author

2025-03-06 01:37 UTC #0 lore raw

Hi all,

Today's linux-next merge of the rdma tree got a conflict in:

drivers/infiniband/sw/rxe/rxe.c

between commit:

8ce2eb9dfac8 ("RDMA/rxe: Fix the failure of ibv_query_device() and ibv_query_device_ex() tests")

from the rdma-fixes tree and commit:

d34d0bdb500e ("RDMA/rxe: Replace netdev dev addr with raw_gid")

from the rdma tree.

I fixed it up (see below) and can carry the fix as necessary. This is now fixed as far as linux-next is concerned, but any non trivial conflicts should be mentioned to your upstream maintainer when your tree is submitted for merging. You may also want to consider cooperating with the maintainer of the conflicting tree to minimise any particularly complex conflicts.

Signature · 74 lines

Cheers,
Stephen Rothwell

diff --cc drivers/infiniband/sw/rxe/rxe.c
index e27478fe9456,4e56a371deb5..000000000000
--- a/drivers/infiniband/sw/rxe/rxe.c
+++ b/drivers/infiniband/sw/rxe/rxe.c
@@@ -72,10 -71,45 +69,39 @@@ static void rxe_init_device_param(struc
  	rxe->attr.max_pkeys			= RXE_MAX_PKEYS;
  	rxe->attr.local_ca_ack_delay		= RXE_LOCAL_CA_ACK_DELAY;
  
 -	ndev = rxe_ib_device_get_netdev(&rxe->ib_dev);
 -	if (!ndev)
 -		return;
 -
+ 	if (ndev->addr_len) {
+ 		memcpy(rxe->raw_gid, ndev->dev_addr,
+ 			min_t(unsigned int, ndev->addr_len, ETH_ALEN));
+ 	} else {
+ 		/*
+ 		 * This device does not have a HW address, but
+ 		 * connection mangagement requires a unique gid.
+ 		 */
+ 		eth_random_addr(rxe->raw_gid);
+ 	}
+ 
  	addrconf_addr_eui48((unsigned char *)&rxe->attr.sys_image_guid,
- 			ndev->dev_addr);
+ 			rxe->raw_gid);
  
 -	dev_put(ndev);
 -
  	rxe->max_ucontext			= RXE_MAX_UCONTEXT;
+ 
+ 	if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) {
+ 		rxe->attr.kernel_cap_flags |= IBK_ON_DEMAND_PAGING;
+ 
+ 		/* IB_ODP_SUPPORT_IMPLICIT is not supported right now. */
+ 		rxe->attr.odp_caps.general_caps |= IB_ODP_SUPPORT;
+ 
+ 		rxe->attr.odp_caps.per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_SEND;
+ 		rxe->attr.odp_caps.per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_RECV;
+ 		rxe->attr.odp_caps.per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV;
+ 
+ 		rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_SEND;
+ 		rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_RECV;
+ 		rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_WRITE;
+ 		rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_READ;
+ 		rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_ATOMIC;
+ 		rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV;
+ 	}
  }
  
  /* initialize port attributes */
@@@ -107,13 -141,18 +133,13 @@@ static void rxe_init_port_param(struct 
  /* initialize port state, note IB convention that HCA ports are always
   * numbered from 1
   */
 -static void rxe_init_ports(struct rxe_dev *rxe)
 +static void rxe_init_ports(struct rxe_dev *rxe, struct net_device *ndev)
  {
  	struct rxe_port *port = &rxe->port;
 -	struct net_device *ndev;
  
  	rxe_init_port_param(port);
 -	ndev = rxe_ib_device_get_netdev(&rxe->ib_dev);
 -	if (!ndev)
 -		return;
  	addrconf_addr_eui48((unsigned char *)&port->port_guid,
- 			    ndev->dev_addr);
+ 			    rxe->raw_gid);
 -	dev_put(ndev);
  	spin_lock_init(&port->port_lock);
  }

Zhu Yanjun

2025-03-06 10:24 UTC #1 lore raw

Hi, Stephen

I am the author of the commit 8ce2eb9dfac8 ("RDMA/rxe: Fix the failure of ibv_query_device() and ibv_query_device_ex() tests").

My replies are inline. Thanks a lot.

drivers/infiniband/sw/rxe/rxe.c +0 -0

diff --cc drivers/infiniband/sw/rxe/rxe.c
index e27478fe9456,4e56a371deb5..000000000000
@@@ -72,10 -71,45 +69,39 @@@ static void rxe_init_device_param(struc

rxe->attr.max_pkeys = RXE_MAX_PKEYS; rxe->attr.local_ca_ack_delay = RXE_LOCAL_CA_ACK_DELAY;

-    ndev = rxe_ib_device_get_netdev(&rxe->ib_dev);
-    if (!ndev)
-        return;
-
+     if (ndev->addr_len) {
+         memcpy(rxe->raw_gid, ndev->dev_addr,
+             min_t(unsigned int, ndev->addr_len, ETH_ALEN));
+     } else {
+         /*
+         * This device does not have a HW address, but
+         * connection mangagement requires a unique gid.
+         */
+         eth_random_addr(rxe->raw_gid);
+     }
+
    addrconf_addr_eui48((unsigned char *)&rxe->attr.sys_image_guid,
-             ndev->dev_addr);
+             rxe->raw_gid);

- dev_put(ndev);
-

------The above snippet is fine with me.

    rxe->max_ucontext            = RXE_MAX_UCONTEXT;
+
+     if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) {
+         rxe->attr.kernel_cap_flags |= IBK_ON_DEMAND_PAGING; +
+         /* IB_ODP_SUPPORT_IMPLICIT is not supported right now. */
+         rxe->attr.odp_caps.general_caps |= IB_ODP_SUPPORT; +

+         rxe->attr.odp_caps.per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_SEND;
+         rxe->attr.odp_caps.per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_RECV;
+         rxe->attr.odp_caps.per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV;
+
+         rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_SEND;
+         rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_RECV;
+         rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_WRITE;
+         rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_READ;
+         rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_ATOMIC;
+         rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV;

+ }
}

---------The above snippet is from on_demand paging commits. I do not why it is involved in this commit.

---------It should appear in on_demand_paging patchset.

/* initialize port attributes */
@@@ -107,13 -141,18 +133,13 @@@ static void rxe_init_port_param(struct /* initialize port state, note IB convention that HCA ports are always
* numbered from 1
*/
-static void rxe_init_ports(struct rxe_dev *rxe)
+static void rxe_init_ports(struct rxe_dev *rxe, struct net_device *ndev)

----------In this snippet, the variable "struct net_device *ndev" is not used in the function rxe_init_ports, it should be removed.

----------Thanks a lot.

Zhu Yanjun

{
struct rxe_port *port = &rxe->port;
- struct net_device *ndev;

    rxe_init_port_param(port);
-    ndev = rxe_ib_device_get_netdev(&rxe->ib_dev); -    if (!ndev)
-        return;
    addrconf_addr_eui48((unsigned char *)&port->port_guid,
-                 ndev->dev_addr);
+                 rxe->raw_gid);
-    dev_put(ndev); spin_lock_init(&port->port_lock);
}

Quoted 23 lines · Stephen Rothwell

Hi all,
Today's linux-next merge of the rdma tree got a conflict in:
drivers/infiniband/sw/rxe/rxe.c
between commit:
8ce2eb9dfac8 ("RDMA/rxe: Fix the failure of ibv_query_device() and ibv_query_device_ex() tests")
from the rdma-fixes tree and commit:
d34d0bdb500e ("RDMA/rxe: Replace netdev dev addr with raw_gid")
from the rdma tree.
I fixed it up (see below) and can carry the fix as necessary. This is now fixed as far as linux-next is concerned, but any non trivial conflicts should be mentioned to your upstream maintainer when your tree is submitted for merging. You may also want to consider cooperating with the maintainer of the conflicting tree to minimise any particularly complex conflicts.

Signature · 2 lines

Best Regards,
Yanjun.Zhu

Zhu Yanjun

2025-03-06 11:35 UTC #2 lore raw

Hi, Stephen

I git clone the latest source code from linux-next.

The head commit is 565351ae7e0c Add linux-next specific files for 20250306

The following should be applied. In linux-next, because raw_gid is used, in the function rxe_init_ports, ndev is no longer used.

So this variable ndev should be removed.

drivers/infiniband/sw/rxe/rxe.c +0 -0

index b248c68bf9b1..c83e2cf82748 100644
@@ -133,7 +133,7 @@ static void rxe_init_port_param(struct rxe_port *port)

/* initialize port state, note IB convention that HCA ports are always
* numbered from 1
*/
-static void rxe_init_ports(struct rxe_dev *rxe, struct net_device *ndev) +static void rxe_init_ports(struct rxe_dev *rxe)
{
struct rxe_port *port = &rxe->port;

@@ -162,7 +162,7 @@ static void rxe_init(struct rxe_dev *rxe, struct net_device *ndev) /* init default device parameters */
rxe_init_device_param(rxe, ndev);

-       rxe_init_ports(rxe, ndev);
+       rxe_init_ports(rxe);
        rxe_init_pools(rxe);

/* init pending mmap list */

Thanks a lot.

Zhu Yanjun

Quoted 138 lines · Zhu Yanjun

Hi, Stephen
I am the author of the commit 8ce2eb9dfac8 ("RDMA/rxe: Fix the failure of ibv_query_device() and ibv_query_device_ex() tests").
My replies are inline. Thanks a lot.
"
drivers/infiniband/sw/rxe/rxe.c +0 -0
diff --cc drivers/infiniband/sw/rxe/rxe.c
index e27478fe9456,4e56a371deb5..000000000000
@@@ -72,10 -71,45 +69,39 @@@ static void rxe_init_device_param(struc
    rxe->attr.max_pkeys            = RXE_MAX_PKEYS; rxe->attr.local_ca_ack_delay        = RXE_LOCAL_CA_ACK_DELAY;
-    ndev = rxe_ib_device_get_netdev(&rxe->ib_dev);
-    if (!ndev)
-        return;
-
+     if (ndev->addr_len) {
+         memcpy(rxe->raw_gid, ndev->dev_addr,
+             min_t(unsigned int, ndev->addr_len, ETH_ALEN));
+     } else {
+         /*
+         * This device does not have a HW address, but
+         * connection mangagement requires a unique gid.
+         */
+         eth_random_addr(rxe->raw_gid);
+     }
+
    addrconf_addr_eui48((unsigned char *)&rxe->attr.sys_image_guid,
-             ndev->dev_addr);
+             rxe->raw_gid);
-    dev_put(ndev);
-
------The above snippet is fine with me.
    rxe->max_ucontext            = RXE_MAX_UCONTEXT;
drivers/infiniband/sw/rxe/rxe.c +8 -0
+
+     if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) {
+         rxe->attr.kernel_cap_flags |= IBK_ON_DEMAND_PAGING;
+
+         /* IB_ODP_SUPPORT_IMPLICIT is not supported right now. */
+         rxe->attr.odp_caps.general_caps |= IB_ODP_SUPPORT;
+
+         rxe->attr.odp_caps.per_transport_caps.ud_odp_caps |= 
IB_ODP_SUPPORT_SEND;
+         rxe->attr.odp_caps.per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_RECV;
+         rxe->attr.odp_caps.per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV;
+
+         rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_SEND;
+         rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_RECV;
+         rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_WRITE;
+         rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_READ;
+         rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_ATOMIC;
+         rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV;
+     }
}
---------The above snippet is from on_demand paging commits. I do not why it is involved in this commit.
---------It should appear in on_demand_paging patchset.
/* initialize port attributes */
drivers/infiniband/sw/rxe/rxe.c +0 -0
@@@ -107,13 -141,18 +133,13 @@@ static void rxe_init_port_param(struct
/* initialize port state, note IB convention that HCA ports are always
   * numbered from 1
   */
-static void rxe_init_ports(struct rxe_dev *rxe)
+static void rxe_init_ports(struct rxe_dev *rxe, struct net_device *ndev)
----------In this snippet, the variable "struct net_device *ndev" is not used in the function rxe_init_ports, it should be removed.
----------Thanks a lot.
Zhu Yanjun
{
    struct rxe_port *port = &rxe->port;
-    struct net_device *ndev;
    rxe_init_port_param(port);
-    ndev = rxe_ib_device_get_netdev(&rxe->ib_dev); -    if (!ndev)
-        return;
    addrconf_addr_eui48((unsigned char *)&port->port_guid,
-                 ndev->dev_addr);
+                 rxe->raw_gid);
-    dev_put(ndev); spin_lock_init(&port->port_lock);
}
"
Quoted 23 lines · Stephen Rothwell
Hi all,
Today's linux-next merge of the rdma tree got a conflict in:
   drivers/infiniband/sw/rxe/rxe.c
between commit:
   8ce2eb9dfac8 ("RDMA/rxe: Fix the failure of ibv_query_device() and ibv_query_device_ex() tests")
from the rdma-fixes tree and commit:
   d34d0bdb500e ("RDMA/rxe: Replace netdev dev addr with raw_gid")
from the rdma tree.
I fixed it up (see below) and can carry the fix as necessary. This is now fixed as far as linux-next is concerned, but any non trivial conflicts should be mentioned to your upstream maintainer when your tree is submitted for merging. You may also want to consider cooperating with the maintainer of the conflicting tree to minimise any particularly complex conflicts.

Signature · 2 lines

Best Regards,
Yanjun.Zhu

A message that is not in the archive (sent off-list or not yet received): [email protected]

Stephen Rothwell

author

2025-03-06 20:33 UTC #4 lore raw

Hi Zhu,

Quoted 24 lines · On Thu, 6 Mar 2025 11:19:24 +0100 Zhu Yanjun <[email protected]> wrote:

Quoted 11 lines

   	rxe->max_ucontext			= RXE_MAX_UCONTEXT;

(quoted diff) +9 -0

+
+ 	if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) {
+ 		rxe->attr.kernel_cap_flags |= IBK_ON_DEMAND_PAGING;
+
+ 		/* IB_ODP_SUPPORT_IMPLICIT is not supported right now. */
+ 		rxe->attr.odp_caps.general_caps |= IB_ODP_SUPPORT;
+
+ 		rxe->attr.odp_caps.per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_SEND;
+ 		rxe->attr.odp_caps.per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_RECV;

Review

Quoted 10 lines

(quoted diff) +9 -0

+ 		rxe->attr.odp_caps.per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV;
+
+ 		rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_SEND;
+ 		rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_RECV;
+ 		rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_WRITE;
+ 		rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_READ;
+ 		rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_ATOMIC;
+ 		rxe->attr.odp_caps.per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV;
+ 	}
   }

The above snippet is from on_demand paging commits. I do not why it is involved in this commit.

Quoted 1 line

It should appear in on_demand_paging patchset.

That is not part of the conflict, but the change is there on one side of the merge.

Quoted 9 lines

Review

Quoted 7 lines

   /* initialize port attributes */

(quoted diff) +0 -0

@@@ -107,13 -141,18 +133,13 @@@ static void rxe_init_port_param(struct
   /* initialize port state, note IB convention that HCA ports are always
    * numbered from 1
    */
  -static void rxe_init_ports(struct rxe_dev *rxe)
  +static void rxe_init_ports(struct rxe_dev *rxe, struct net_device *ndev)

In this snippet, the variable "struct net_device *ndev" is not used in the function rxe_init_ports, it should be removed.

I have added that to my resolution.

Signature · 2 lines

Cheers,
Stephen Rothwell