diff mbox series

[net,v1] i40e: Fix queues reservation for XDP

Message ID 20211126101122.25085-1-mateusz.palczewski@intel.com
State Accepted
Delegated to: Anthony Nguyen
Headers show
Series [net,v1] i40e: Fix queues reservation for XDP | expand

Commit Message

Mateusz Palczewski Nov. 26, 2021, 10:11 a.m. UTC 
From: Sylwester Dziedziuch <sylwesterx.dziedziuch@intel.com>

When XDP was configured on a system with large number of CPUs
and X722 NIC there was a call trace with NULL pointer dereference.

i40e 0000:87:00.0: failed to get tracking for 256 queues for VSI 0 err -12
i40e 0000:87:00.0: setup of MAIN VSI failed

BUG: kernel NULL pointer dereference, address: 0000000000000000
RIP: 0010:i40e_xdp+0xea/0x1b0 [i40e]
Call Trace:
? i40e_reconfig_rss_queues+0x130/0x130 [i40e]
dev_xdp_install+0x61/0xe0
dev_xdp_attach+0x18a/0x4c0
dev_change_xdp_fd+0x1e6/0x220
do_setlink+0x616/0x1030
? ahci_port_stop+0x80/0x80
? ata_qc_issue+0x107/0x1e0
? lock_timer_base+0x61/0x80
? __mod_timer+0x202/0x380
rtnl_setlink+0xe5/0x170
? bpf_lsm_binder_transaction+0x10/0x10
? security_capable+0x36/0x50
rtnetlink_rcv_msg+0x121/0x350
? rtnl_calcit.isra.0+0x100/0x100
netlink_rcv_skb+0x50/0xf0
netlink_unicast+0x1d3/0x2a0
netlink_sendmsg+0x22a/0x440
sock_sendmsg+0x5e/0x60
__sys_sendto+0xf0/0x160
? __sys_getsockname+0x7e/0xc0
? _copy_from_user+0x3c/0x80
? __sys_setsockopt+0xc8/0x1a0
__x64_sys_sendto+0x20/0x30
do_syscall_64+0x33/0x40
entry_SYSCALL_64_after_hwframe+0x44/0xae
RIP: 0033:0x7f83fa7a39e0

This was caused by PF queue pile fragmentation due to
flow director VSI queue being placed right after main VSI.
Because of this main VSI was not able to resize its
queue allocation for XDP resulting in no queues allocated
for main VSI when XDP was turned on.

Fix this by always allocating last queue in PF queue pile
for a flow director VSI.

Fixes: 41c445ff0f48 ("i40e: main driver core")
Signed-off-by: Sylwester Dziedziuch <sylwesterx.dziedziuch@intel.com>
Signed-off-by: Mateusz Palczewski <mateusz.palczewski@intel.com>
---
 drivers/net/ethernet/intel/i40e/i40e_main.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

Comments

Paul Menzel Nov. 26, 2021, 2:47 p.m. UTC | #1 
Dear Maciej,


Am 26.11.21 um 16:55 schrieb Maciej Fijalkowski:

[…]

The date of your message is again from the future. :( Can the Intel 
office in Poland please fix their systems. It’s happening again and again.


Kind regards,

Paul
Maciej Fijalkowski Nov. 26, 2021, 3:55 p.m. UTC | #2 
On Fri, Nov 26, 2021 at 11:11:22AM +0100, Mateusz Palczewski wrote:
> From: Sylwester Dziedziuch <sylwesterx.dziedziuch@intel.com>
> 
> When XDP was configured on a system with large number of CPUs
> and X722 NIC there was a call trace with NULL pointer dereference.
> 
> i40e 0000:87:00.0: failed to get tracking for 256 queues for VSI 0 err -12
> i40e 0000:87:00.0: setup of MAIN VSI failed
> 
> BUG: kernel NULL pointer dereference, address: 0000000000000000
> RIP: 0010:i40e_xdp+0xea/0x1b0 [i40e]
> Call Trace:
> ? i40e_reconfig_rss_queues+0x130/0x130 [i40e]
> dev_xdp_install+0x61/0xe0
> dev_xdp_attach+0x18a/0x4c0
> dev_change_xdp_fd+0x1e6/0x220
> do_setlink+0x616/0x1030
> ? ahci_port_stop+0x80/0x80
> ? ata_qc_issue+0x107/0x1e0
> ? lock_timer_base+0x61/0x80
> ? __mod_timer+0x202/0x380
> rtnl_setlink+0xe5/0x170
> ? bpf_lsm_binder_transaction+0x10/0x10
> ? security_capable+0x36/0x50
> rtnetlink_rcv_msg+0x121/0x350
> ? rtnl_calcit.isra.0+0x100/0x100
> netlink_rcv_skb+0x50/0xf0
> netlink_unicast+0x1d3/0x2a0
> netlink_sendmsg+0x22a/0x440
> sock_sendmsg+0x5e/0x60
> __sys_sendto+0xf0/0x160
> ? __sys_getsockname+0x7e/0xc0
> ? _copy_from_user+0x3c/0x80
> ? __sys_setsockopt+0xc8/0x1a0
> __x64_sys_sendto+0x20/0x30
> do_syscall_64+0x33/0x40
> entry_SYSCALL_64_after_hwframe+0x44/0xae
> RIP: 0033:0x7f83fa7a39e0
> 
> This was caused by PF queue pile fragmentation due to
> flow director VSI queue being placed right after main VSI.
> Because of this main VSI was not able to resize its
> queue allocation for XDP resulting in no queues allocated
> for main VSI when XDP was turned on.
> 
> Fix this by always allocating last queue in PF queue pile
> for a flow director VSI.
> 
> Fixes: 41c445ff0f48 ("i40e: main driver core")

I'd also add:
Fixes: 74608d17fe29 ("i40e: add support for XDP_TX action")

> Signed-off-by: Sylwester Dziedziuch <sylwesterx.dziedziuch@intel.com>
> Signed-off-by: Mateusz Palczewski <mateusz.palczewski@intel.com>

Reviewed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>

> ---
>  drivers/net/ethernet/intel/i40e/i40e_main.c | 14 ++++++++++++++
>  1 file changed, 14 insertions(+)
> 
> diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c b/drivers/net/ethernet/intel/i40e/i40e_main.c
> index 86fb795..9c612b1 100644
> --- a/drivers/net/ethernet/intel/i40e/i40e_main.c
> +++ b/drivers/net/ethernet/intel/i40e/i40e_main.c
> @@ -192,6 +192,20 @@ static int i40e_get_lump(struct i40e_pf *pf, struct i40e_lump_tracking *pile,
>  		return -EINVAL;
>  	}
>  
> +	/* Allocate last queue in the pile for FDIR VSI queue
> +	 * so it doesn't fragment the qp_pile
> +	 */
> +	if (pile == pf->qp_pile && pf->vsi[id]->type == I40E_VSI_FDIR) {
> +		if (pile->list[pile->num_entries - 1] & I40E_PILE_VALID_BIT) {
> +			dev_err(&pf->pdev->dev,
> +				"Cannot allocate queue %d for I40E_VSI_FDIR\n",
> +				pile->num_entries - 1);
> +			return -ENOMEM;
> +		}
> +		pile->list[pile->num_entries - 1] = id | I40E_PILE_VALID_BIT;
> +		return pile->num_entries - 1;
> +	}
> +
>  	i = 0;
>  	while (i < pile->num_entries) {
>  		/* skip already allocated entries */
> -- 
> 2.27.0
> 
> _______________________________________________
> Intel-wired-lan mailing list
> Intel-wired-lan@osuosl.org
> https://lists.osuosl.org/mailman/listinfo/intel-wired-lan
Bhandare, KiranX Jan. 5, 2022, 6:14 a.m. UTC | #3 
> -----Original Message-----
> From: Intel-wired-lan <intel-wired-lan-bounces@osuosl.org> On Behalf Of
> Mateusz Palczewski
> Sent: Friday, November 26, 2021 3:41 PM
> To: intel-wired-lan@lists.osuosl.org
> Cc: Dziedziuch, SylwesterX <sylwesterx.dziedziuch@intel.com>; Palczewski,
> Mateusz <mateusz.palczewski@intel.com>
> Subject: [Intel-wired-lan] [PATCH net v1] i40e: Fix queues reservation for XDP
> 
> From: Sylwester Dziedziuch <sylwesterx.dziedziuch@intel.com>
> 
> When XDP was configured on a system with large number of CPUs and X722
> NIC there was a call trace with NULL pointer dereference.
> 
> i40e 0000:87:00.0: failed to get tracking for 256 queues for VSI 0 err -12 i40e
> 0000:87:00.0: setup of MAIN VSI failed
> 
> BUG: kernel NULL pointer dereference, address: 0000000000000000
> RIP: 0010:i40e_xdp+0xea/0x1b0 [i40e]
> Call Trace:
> ? i40e_reconfig_rss_queues+0x130/0x130 [i40e]
> dev_xdp_install+0x61/0xe0
> dev_xdp_attach+0x18a/0x4c0
> dev_change_xdp_fd+0x1e6/0x220
> do_setlink+0x616/0x1030
> ? ahci_port_stop+0x80/0x80
> ? ata_qc_issue+0x107/0x1e0
> ? lock_timer_base+0x61/0x80
> ? __mod_timer+0x202/0x380
> rtnl_setlink+0xe5/0x170
> ? bpf_lsm_binder_transaction+0x10/0x10
> ? security_capable+0x36/0x50
> rtnetlink_rcv_msg+0x121/0x350
> ? rtnl_calcit.isra.0+0x100/0x100
> netlink_rcv_skb+0x50/0xf0
> netlink_unicast+0x1d3/0x2a0
> netlink_sendmsg+0x22a/0x440
> sock_sendmsg+0x5e/0x60
> __sys_sendto+0xf0/0x160
> ? __sys_getsockname+0x7e/0xc0
> ? _copy_from_user+0x3c/0x80
> ? __sys_setsockopt+0xc8/0x1a0
> __x64_sys_sendto+0x20/0x30
> do_syscall_64+0x33/0x40
> entry_SYSCALL_64_after_hwframe+0x44/0xae
> RIP: 0033:0x7f83fa7a39e0
> 
> This was caused by PF queue pile fragmentation due to flow director VSI
> queue being placed right after main VSI.
> Because of this main VSI was not able to resize its queue allocation for XDP
> resulting in no queues allocated for main VSI when XDP was turned on.
> 
> Fix this by always allocating last queue in PF queue pile for a flow director
> VSI.
> 
> Fixes: 41c445ff0f48 ("i40e: main driver core")
> Signed-off-by: Sylwester Dziedziuch <sylwesterx.dziedziuch@intel.com>
> Signed-off-by: Mateusz Palczewski <mateusz.palczewski@intel.com>
> ---
>  drivers/net/ethernet/intel/i40e/i40e_main.c | 14 ++++++++++++++
>  1 file changed, 14 insertions(+)
> 

Tested-by: Kiran Bhandare <kiranx.bhandare@intel.com>  A Contingent Worker at Intel
diff mbox series

Patch

diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c b/drivers/net/ethernet/intel/i40e/i40e_main.c
index 86fb795..9c612b1 100644
--- a/drivers/net/ethernet/intel/i40e/i40e_main.c
+++ b/drivers/net/ethernet/intel/i40e/i40e_main.c
@@ -192,6 +192,20 @@  static int i40e_get_lump(struct i40e_pf *pf, struct i40e_lump_tracking *pile,
 		return -EINVAL;
 	}
 
+	/* Allocate last queue in the pile for FDIR VSI queue
+	 * so it doesn't fragment the qp_pile
+	 */
+	if (pile == pf->qp_pile && pf->vsi[id]->type == I40E_VSI_FDIR) {
+		if (pile->list[pile->num_entries - 1] & I40E_PILE_VALID_BIT) {
+			dev_err(&pf->pdev->dev,
+				"Cannot allocate queue %d for I40E_VSI_FDIR\n",
+				pile->num_entries - 1);
+			return -ENOMEM;
+		}
+		pile->list[pile->num_entries - 1] = id | I40E_PILE_VALID_BIT;
+		return pile->num_entries - 1;
+	}
+
 	i = 0;
 	while (i < pile->num_entries) {
 		/* skip already allocated entries */