Skip to contentSkip to conversation

RDMA/siw: Remove unwanted WARN_ON in siw_cm_llp_data_ready()

PATCH In review Krishnamraju Eraparaju 3 participants 7 messages Feb 7, 2020 – Feb 11, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 RDMA/siw: Remove unwanted WARN_ON in siw_cm_llp_data_ready()

Warnings like below can fill up the dmesg while disconnecting RDMA connections.
Hence, removing the unwanted WARN_ON.

[72103.557612] WARNING: CPU: 6 PID: 0 at
drivers/infiniband/sw/siw/siw_cm.c:1229 siw_cm_llp_data_ready+0xc1/0xd0
[siw]
[72103.557677] RIP: 0010:siw_cm_llp_data_ready+0xc1/0xd0 [siw]
[72103.557693] Call Trace:
[72103.557711]  tcp_data_queue+0x226/0xb40
[72103.557714]  tcp_rcv_established+0x220/0x620
[72103.557720]  tcp_v4_do_rcv+0x12a/0x1e0
[72103.557722]  tcp_v4_rcv+0xb05/0xc00
[72103.557728]  ip_local_deliver_finish+0x69/0x210
[72103.557730]  ip_local_deliver+0x6b/0xe0
[72103.557735]  ip_rcv+0x273/0x362
[72103.557740]  __netif_receive_skb_core+0xb35/0xc30
[72103.557752]  netif_receive_skb_internal+0x3d/0xb0
[72103.557754]  napi_gro_frags+0x13b/0x200
[72103.557788]  t4_ethrx_handler+0x433/0x7d0 [cxgb4]
[72103.557800]  process_responses+0x318/0x580 [cxgb4]
[72103.557820]  napi_rx_handler+0x14/0x100 [cxgb4]
[72103.557822]  net_rx_action+0x149/0x3b0
[72103.557826]  __do_softirq+0xe3/0x30a
[72103.557831]  irq_exit+0x100/0x110
[72103.557834]  do_IRQ+0x7f/0xe0
[72103.557837]  common_interrupt+0xf/0xf
---
 drivers/infiniband/sw/siw/siw_cm.c | 1 -
 1 file changed, 1 deletion(-)
drivers/infiniband/sw/siw/siw_cm.c +0 -1
index 3bccfef40e7e..bcbc54998ace 100644
@@ -1226,7 +1226,6 @@ static void siw_cm_llp_data_ready(struct sock *sk)
 
 	cep = sk_to_cep(sk);
 	if (!cep) {
-		WARN_ON(1);
 		goto out;
 	}
 	siw_dbg_cep(cep, "state: %d\n", cep->state);
Signature · 1 line
2.23.0.rc0
#1 lore raw
Quoted 48 lines · Krishnamraju Eraparaju

Warnings like below can fill up the dmesg while disconnecting RDMA connections.
Hence, removing the unwanted WARN_ON.

[72103.557612] WARNING: CPU: 6 PID: 0 at
drivers/infiniband/sw/siw/siw_cm.c:1229 siw_cm_llp_data_ready+0xc1/0xd0
[siw]
[72103.557677] RIP: 0010:siw_cm_llp_data_ready+0xc1/0xd0 [siw]
[72103.557693] Call Trace:
[72103.557711]  tcp_data_queue+0x226/0xb40
[72103.557714]  tcp_rcv_established+0x220/0x620
[72103.557720]  tcp_v4_do_rcv+0x12a/0x1e0
[72103.557722]  tcp_v4_rcv+0xb05/0xc00
[72103.557728]  ip_local_deliver_finish+0x69/0x210
[72103.557730]  ip_local_deliver+0x6b/0xe0
[72103.557735]  ip_rcv+0x273/0x362
[72103.557740]  __netif_receive_skb_core+0xb35/0xc30
[72103.557752]  netif_receive_skb_internal+0x3d/0xb0
[72103.557754]  napi_gro_frags+0x13b/0x200
[72103.557788]  t4_ethrx_handler+0x433/0x7d0 [cxgb4]
[72103.557800]  process_responses+0x318/0x580 [cxgb4]
[72103.557820]  napi_rx_handler+0x14/0x100 [cxgb4]
[72103.557822]  net_rx_action+0x149/0x3b0
[72103.557826]  __do_softirq+0xe3/0x30a
[72103.557831]  irq_exit+0x100/0x110
[72103.557834]  do_IRQ+0x7f/0xe0
[72103.557837]  common_interrupt+0xf/0xf
---
 drivers/infiniband/sw/siw/siw_cm.c | 1 -
 1 file changed, 1 deletion(-)
drivers/infiniband/sw/siw/siw_cm.c +0 -1
index 3bccfef40e7e..bcbc54998ace 100644
@@ -1226,7 +1226,6 @@ static void siw_cm_llp_data_ready(struct sock *sk)
 
 	cep = sk_to_cep(sk);
 	if (!cep) {
-		WARN_ON(1);
 		goto out;
 	}
 	siw_dbg_cep(cep, "state: %d\n", cep->state);
Signature · 1 line
2.23.0.rc0

Please ignore this patch as I mistakenly put "for-review/for-rc/for-rc" in the subject line instead of "for-rc".

#2 lore raw
In reply to Krishnamraju Eraparaju
Quoted 3 lines · Krishnamraju Eraparaju

Warnings like below can fill up the dmesg while disconnecting RDMA connections.
Hence, removing the unwanted WARN_ON.

Please explain why it the code is correct to take this error path. Bernard clearly thought this shouldn't be happening

Jason

#3 lore raw

-----"Jason Gunthorpe" <[email protected]> wrote: -----

Quoted 10 lines

To: "Krishnamraju Eraparaju" <[email protected]>
From: "Jason Gunthorpe" <[email protected]>
Date: 02/07/2020 03:18PM
Cc: [email protected], [email protected], [email protected], [email protected], [email protected]
Subject: [EXTERNAL] Re: [PATCH for-review/for-rc/for-rc] RDMA/siw:
Remove unwanted WARN_ON in siw_cm_llp_data_ready()

On Fri, Feb 07, 2020 at 05:22:09PM +0530, Krishnamraju Eraparaju wrote:

Quoted 10 lines
Quoted 3 lines

Warnings like below can fill up the dmesg while disconnecting RDMA connections.
Hence, removing the unwanted WARN_ON.

Please explain why it the code is correct to take this error path. Bernard clearly thought this shouldn't be happening

Jason

I have to look into it as well. It's maybe on me revising my thoughts.

Thanks,
Bernard.

#4 lore raw
In reply to Jason Gunthorpe
Quoted 9 lines · Jason Gunthorpe
Quoted 3 lines · Krishnamraju Eraparaju

Warnings like below can fill up the dmesg while disconnecting RDMA connections.
Hence, removing the unwanted WARN_ON.

Please explain why it the code is correct to take this error path. Bernard clearly thought this shouldn't be happening

Jason

As part of iSER multipath testcase, target(iw_cxgb4) responds with MPA reject to initiator(SIW) when iw_cxgb4 resources gets exhaused(expected as per testcase), then SIW performs the connection teardown and dissociates 'cep' from tcp socket 'sk'. And if any "data_ready" notifications from TCP stack after this connection teardown will hit WARN_ON() in siw_cm_llp_data_ready().

Bernard, is this WARN_ON() useful to identify any error conditions?

Thanks,
Krishna.

#5 lore raw

-----"Krishnamraju Eraparaju" <[email protected]> wrote: -----

Quoted 31 lines

To: "Jason Gunthorpe" <[email protected]>
From: "Krishnamraju Eraparaju" <[email protected]>
Date: 02/10/2020 07:00PM
Cc: [email protected], [email protected], [email protected], [email protected], [email protected]
Subject: [EXTERNAL] Re: [PATCH for-review/for-rc/for-rc] RDMA/siw:
Remove unwanted WARN_ON in siw_cm_llp_data_ready()

Quoted 1 line · Jason Gunthorpe

On Fri, Feb 07, 2020 at 05:22:09PM +0530, Krishnamraju Eraparaju

Quoted 1 line · wrote:
Quoted 1 line

Warnings like below can fill up the dmesg while disconnecting

RDMA

Quoted 7 lines
Quoted 2 lines

connections.
Hence, removing the unwanted WARN_ON.

Please explain why it the code is correct to take this error path. Bernard clearly thought this shouldn't be happening

Jason

As part of iSER multipath testcase, target(iw_cxgb4) responds with MPA reject
to initiator(SIW) when iw_cxgb4 resources gets exhaused(expected as per
testcase), then SIW performs the connection teardown and dissociates 'cep' from tcp socket 'sk'. And if any "data_ready" notifications from
TCP stack after this connection teardown will hit WARN_ON() in siw_cm_llp_data_ready().

Quoted 2 lines

Bernard, is this WARN_ON() useful to identify any error conditions?

Let me try recreating the issue. I'll come back asap.

Thanks
Bernard.

#6 lore raw
In reply to Krishnamraju Eraparaju

-----"Krishnamraju Eraparaju" <[email protected]> wrote: -----

Quoted 31 lines

To: "Jason Gunthorpe" <[email protected]>
From: "Krishnamraju Eraparaju" <[email protected]>
Date: 02/10/2020 07:00PM
Cc: [email protected], [email protected], [email protected], [email protected], [email protected]
Subject: [EXTERNAL] Re: [PATCH for-review/for-rc/for-rc] RDMA/siw:
Remove unwanted WARN_ON in siw_cm_llp_data_ready()

Quoted 1 line · Jason Gunthorpe

On Fri, Feb 07, 2020 at 05:22:09PM +0530, Krishnamraju Eraparaju

Quoted 1 line · wrote:
Quoted 1 line

Warnings like below can fill up the dmesg while disconnecting

RDMA

Quoted 7 lines
Quoted 2 lines

connections.
Hence, removing the unwanted WARN_ON.

Please explain why it the code is correct to take this error path. Bernard clearly thought this shouldn't be happening

Jason

As part of iSER multipath testcase, target(iw_cxgb4) responds with MPA reject
to initiator(SIW) when iw_cxgb4 resources gets exhaused(expected as per
testcase), then SIW performs the connection teardown and dissociates 'cep' from tcp socket 'sk'. And if any "data_ready" notifications from
TCP stack after this connection teardown will hit WARN_ON() in siw_cm_llp_data_ready().

Quoted 2 lines

Bernard, is this WARN_ON() useful to identify any error conditions?

So, this WARN_ON() is wrong. It can very well happen that the socket already got disassociated from siw, but a sockets data_ready() upcall races with the disassociation procedure. Since we hold the sk_callback_lock while setting or testing sk->sk_user_data, a NULL pointer tells us this socket just got disassociated from siw and we can bail out.
Thanks Krishna for finding that out. Your patch is correct.

Bernard.