Skip to contentSkip to conversation

cxgb4vf: fix crash due to manipulating queues before registration

PATCH In review Casey Leedom 4 participants 7 messages Oct 28, 2010 – Nov 1, 2010

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

PATCH 1/1 cxgb4vf: fix crash due to manipulating queues before registration

Before commit "net: allocate tx queues in register_netdevice" netif_tx_stop_all_queues and related functions could be used between device allocation and registration but now only after registration. cxgb4 has such a call before registration and crashes now. Move it after register_netdev.

---
 drivers/net/cxgb4vf/cxgb4vf_main.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/net/cxgb4vf/cxgb4vf_main.c +1 -1
index 555ecc5..b9d92a5 100644
@@ -2600,7 +2600,6 @@ static int __devinit cxgb4vf_pci_probe(struct pci_dev *pdev,
 		pi->xact_addr_filt = -1;
 		pi->rx_offload = RX_CSO;
 		netif_carrier_off(netdev);
-		netif_tx_stop_all_queues(netdev);
 		netdev->irq = pdev->irq;
 
 		netdev->features = (NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
@@ -2661,6 +2660,7 @@ static int __devinit cxgb4vf_pci_probe(struct pci_dev *pdev,
 			continue;
 		}
 
+		netif_tx_stop_all_queues(netdev);
 		set_bit(pidx, &adapter->registered_device_map);
 	}
 	if (adapter->registered_device_map == 0) {
Signature · 1 line
1.7.0.4
maintainer
#1 lore raw

From: Casey Leedom <[email protected]>
Date: Thu, 28 Oct 2010 13:16:47 -0700

Quoted 7 lines

Before commit "net: allocate tx queues in register_netdevice" netif_tx_stop_all_queues and related functions could be used between device allocation and registration but now only after registration. cxgb4 has such a call before registration and crashes now. Move it after register_netdev.

Why are you manipulating the queue at all here?

The queue state is "don't care" at this point in time, and has no meaning until ->open() is invoked.

#2 lore raw
Quoted 9 lines · David Miller

From: Casey Leedom <[email protected]>
Date: Thu, 28 Oct 2010 13:16:47 -0700

Quoted 6 lines

Before commit "net: allocate tx queues in register_netdevice" netif_tx_stop_all_queues and related functions could be used between device allocation and registration but now only after registration. cxgb4 has such a call before registration and crashes now. Move it after register_netdev.

Quoted 6 lines
Quoted 1 line

Why are you manipulating the queue at all here?

The queue state is "don't care" at this point in time, and has no meaning until ->open() is invoked.

True. This driver was modeled on cxgb4, we wanted to make sure that the cxgb4 crash fix is ported to cxgb4vf. Let me sync up with the other Chelsio maintainers.

Casey

#3 lore raw
In reply to David Miller
Quoted 6 lines
Quoted 6 lines

Before commit "net: allocate tx queues in register_netdevice" netif_tx_stop_all_queues and related functions could be used between device allocation and registration but now only after registration. cxgb4 has such a call before registration and crashes now. Move it after register_netdev.

Quoted 6 lines
Quoted 1 line

Why are you manipulating the queue at all here?

The queue state is "don't care" at this point in time, and has no meaning until ->open() is invoked.

The reason for the call at its initial pre register_netdev location was an attempt to make the queues have the same state at the time of each open. Without it at the first open __QUEUE_STATE_XOFF is not set due to 0-initialization but at subsequent opens it is set because close stops the queues. I agree however that this disparity is not functionally significant. Further, I believe moving the call after register_netdev is buggy as open can be called after registration and it can clash with the queue stopping. It seems then that these netif_tx_stop_all_queues calls have to go now.

maintainer
#4 lore raw

From: "Dimitrios Michailidis" <[email protected]>
Date: Fri, 29 Oct 2010 00:36:22 -0700

Quoted 4 lines

Further, I believe moving the call after register_netdev is buggy as open can be called after registration and it can clash with the queue stopping. It seems then that these netif_tx_stop_all_queues calls have to go now.

This is a good explanation of why no driver should be touching the queue state before the first ->open() call.

#5 lore raw
Quoted 9 lines · David Miller

From: "Dimitrios Michailidis"<[email protected]>
Date: Fri, 29 Oct 2010 00:36:22 -0700

Quoted 4 lines

Further, I believe moving the call after register_netdev is buggy as open can be called after registration and it can clash with the queue stopping. It seems then that these netif_tx_stop_all_queues calls have to go now.

This is a good explanation of why no driver should be touching the queue state before the first ->open() call.

I'm sending a series of 3 patches fixing cxgb3, cxgb4, cxgb4vf.

Cheers,
Divy

maintainer
#6 lore raw

From: Divy Le Ray <[email protected]>
Date: Mon, 01 Nov 2010 14:02:56 -0700

Quoted 12 lines · David Miller
Quoted 9 lines

From: "Dimitrios Michailidis"<[email protected]>
Date: Fri, 29 Oct 2010 00:36:22 -0700

Quoted 4 lines

Further, I believe moving the call after register_netdev is buggy as open can be called after registration and it can clash with the queue stopping. It seems then that these netif_tx_stop_all_queues calls have to go now.

This is a good explanation of why no driver should be touching the queue state before the first ->open() call.

I'm sending a series of 3 patches fixing cxgb3, cxgb4, cxgb4vf.

Thank you.