diff mbox

Atheros Communications Inc. AR8121/AR8113/AR8114 Gigabit or Fast Ethernet (rev b0) 1.0.0.7 md5/sha1 corrupted using NFS and samba (updated) Version 2

Message ID 20130328031932.GC2176@order.stressinduktion.org
State RFC, archived
Delegated to: David Miller
Headers show

Commit Message

Hannes Frederic Sowa March 28, 2013, 3:19 a.m. UTC 
On Thu, Mar 28, 2013 at 01:17:53AM +0000, Huang, Xiong wrote:
> > 
> > We will try to go with RXQ_CTRL_PBA_ALIGN_256 in REG_RXQ_CTRL next.
> > Just have to check the ring resources are initialized correctly.
> > 
> > Thanks,
> 
> 
> ALIGN_256 require the RXF page 256 byte alignment, you should also revise the memory allocation function.

Could you have a look at the following diff? It does not work either. Perhaps
something is wrong in the calculations (but there are no seq number mismatches).

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Huang, Xiong March 28, 2013, 3:25 a.m. UTC | #1 
Please don't revise following code:

			rx_page->read_offset +=
				(((u32)((prrs->word1 >> RRS_PKT_SIZE_SHIFT) &
				RRS_PKT_SIZE_MASK) +
				sizeof(struct atl1e_recv_ret_status) + 31) &
						0xFFFFFFE0);

Here 32bytes alignment is no matter with Page alignment.

Thanks
Xiong


> -----Original Message-----

> From: Hannes Frederic Sowa [mailto:hannes@stressinduktion.org]

> Sent: Thursday, March 28, 2013 11:20 AM

> To: Huang, Xiong

> Cc: Sven Hartge; netdev@vger.kernel.org

> Subject: Re: Atheros Communications Inc. AR8121/AR8113/AR8114 Gigabit or

> Fast Ethernet (rev b0) 1.0.0.7 md5/sha1 corrupted using NFS and samba

> (updated) Version 2

> 

> On Thu, Mar 28, 2013 at 01:17:53AM +0000, Huang, Xiong wrote:

> > >

> > > We will try to go with RXQ_CTRL_PBA_ALIGN_256 in REG_RXQ_CTRL next.

> > > Just have to check the ring resources are initialized correctly.

> > >

> > > Thanks,

> >

> >

> > ALIGN_256 require the RXF page 256 byte alignment, you should also revise

> the memory allocation function.

> 

> Could you have a look at the following diff? It does not work either. Perhaps

> something is wrong in the calculations (but there are no seq number

> mismatches).

> 

> diff --git a/drivers/net/ethernet/atheros/atl1e/atl1e_main.c

> b/drivers/net/ethernet/atheros/atl1e/atl1e_main.c

> index e1f1b2a..5264810 100644

> --- a/drivers/net/ethernet/atheros/atl1e/atl1e_main.c

> +++ b/drivers/net/ethernet/atheros/atl1e/atl1e_main.c

> @@ -697,7 +697,7 @@ static void atl1e_cal_ring_size(struct atl1e_adapter

> *adapter, u32 *ring_size)

>  		     sizeof(struct atl1e_tpd_desc) + 7

>  			/* tx ring, qword align */

>  		     + adapter->rx_ring.real_page_size *

> AT_PAGE_NUM_PER_QUEUE *

> -			adapter->num_rx_queues + 31

> +			adapter->num_rx_queues + 255

>  			/* rx ring,  32 bytes align */

>  		     + (1 + AT_PAGE_NUM_PER_QUEUE * adapter-

> >num_rx_queues) *

>  			sizeof(u32) + 3));

> @@ -714,7 +714,7 @@ static void atl1e_init_ring_resources(struct

> atl1e_adapter *adapter)

>  				 + adapter->hw.max_frame_size

>  				 + ETH_HLEN + VLAN_HLEN

>  				 + ETH_FCS_LEN;

> -	rx_ring->real_page_size = roundup(rx_ring->real_page_size, 32);

> +	rx_ring->real_page_size = roundup(rx_ring->real_page_size, 256);

>  	atl1e_cal_ring_size(adapter, &adapter->ring_size);

> 

>  	adapter->ring_vir_addr = NULL;

> @@ -825,7 +825,7 @@ static int atl1e_setup_ring_resources(struct

> atl1e_adapter *adapter)

> 

>  	/* Init RXF-Pages */

>  	offset += (sizeof(struct atl1e_tpd_desc) * tx_ring->count);

> -	offset = roundup(offset, 32);

> +	offset = roundup(offset, 256);

> 

>  	for (i = 0; i < adapter->num_rx_queues; i++) {

>  		for (j = 0; j < AT_PAGE_NUM_PER_QUEUE; j++) { @@ -1003,7

> +1003,7 @@ static inline void atl1e_configure_rx(struct atl1e_adapter

> *adapter)

>  		rxq_ctrl_data |=

>  			(RXQ_CTRL_HASH_ENABLE |

> RXQ_CTRL_RSS_MODE_MQUESINT);

> 

> -	rxq_ctrl_data |= RXQ_CTRL_IPV6_XSUM_VERIFY_EN |

> RXQ_CTRL_PBA_ALIGN_32 |

> +	rxq_ctrl_data |= RXQ_CTRL_IPV6_XSUM_VERIFY_EN |

> RXQ_CTRL_PBA_ALIGN_256

> +|

>  			 RXQ_CTRL_CUT_THRU_EN | RXQ_CTRL_EN;

> 

>  	AT_WRITE_REG(hw, REG_RXQ_CTRL, rxq_ctrl_data); @@ -1444,8

> +1444,8 @@ skip_pkt:

>  			rx_page->read_offset +=

>  				(((u32)((prrs->word1 >>

> RRS_PKT_SIZE_SHIFT) &

>  				RRS_PKT_SIZE_MASK) +

> -				sizeof(struct atl1e_recv_ret_status) + 31) &

> -						0xFFFFFFE0);

> +				sizeof(struct atl1e_recv_ret_status) + 255) &

> +						0xFFFFFF00);

> 

>  			if (rx_page->read_offset >= rx_ring->page_size) {

>  				/* mark this page clean */
Hannes Frederic Sowa March 28, 2013, 3:29 a.m. UTC | #2 
On Thu, Mar 28, 2013 at 03:25:29AM +0000, Huang, Xiong wrote:
> Please don't revise following code:
> 
> 			rx_page->read_offset +=
> 				(((u32)((prrs->word1 >> RRS_PKT_SIZE_SHIFT) &
> 				RRS_PKT_SIZE_MASK) +
> 				sizeof(struct atl1e_recv_ret_status) + 31) &
> 						0xFFFFFFE0);
> 
> Here 32bytes alignment is no matter with Page alignment.

If I revert this change I get sequence errors:

[  781.449864] ATL1E 0000:04:00.0: irq 45 for MSI/MSI-X
[  781.454634] IPv6: ADDRCONF(NETDEV_UP): p33p1: link is not ready
[  782.893744] ATL1E 0000:04:00.0 p33p1: NIC Link is Up <100 Mbps Full Duplex>
[  782.893797] IPv6: ADDRCONF(NETDEV_CHANGE): p33p1: link becomes ready
[  783.194443] ATL1E 0000:04:00.0 p33p1: rx sequence number error (rx=0) (expect=1)
[  783.204497] ATL1E 0000:04:00.0 p33p1: NIC Link is Up <100 Mbps Full Duplex>
[  783.225336] ATL1E 0000:04:00.0 p33p1: rx sequence number error (rx=0) (expect=1)
[  783.236636] ATL1E 0000:04:00.0 p33p1: NIC Link is Up <100 Mbps Full Duplex>
[  783.462652] ATL1E 0000:04:00.0 p33p1: rx sequence number error (rx=0) (expect=1)
[  783.472530] ATL1E 0000:04:00.0 p33p1: NIC Link is Up <100 Mbps Full Duplex>
[  784.038976] ATL1E 0000:04:00.0 p33p1: rx sequence number error (rx=0) (expect=1)
[  784.047354] ATL1E 0000:04:00.0 p33p1: NIC Link is Up <100 Mbps Full Duplex>
...

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Huang, Xiong March 28, 2013, 3:39 a.m. UTC | #3 
Dump every packet length & read_offset value,  compare them.

> -----Original Message-----

> From: Hannes Frederic Sowa [mailto:hannes@stressinduktion.org]

> Sent: Thursday, March 28, 2013 11:30 AM

> To: Huang, Xiong

> Cc: Sven Hartge; netdev@vger.kernel.org

> Subject: Re: Atheros Communications Inc. AR8121/AR8113/AR8114 Gigabit or

> Fast Ethernet (rev b0) 1.0.0.7 md5/sha1 corrupted using NFS and samba

> (updated) Version 2

> 

> On Thu, Mar 28, 2013 at 03:25:29AM +0000, Huang, Xiong wrote:

> > Please don't revise following code:

> >

> > 			rx_page->read_offset +=

> > 				(((u32)((prrs->word1 >>

> RRS_PKT_SIZE_SHIFT) &

> > 				RRS_PKT_SIZE_MASK) +

> > 				sizeof(struct atl1e_recv_ret_status) + 31) &

> > 						0xFFFFFFE0);

> >

> > Here 32bytes alignment is no matter with Page alignment.

> 

> If I revert this change I get sequence errors:

> 

> [  781.449864] ATL1E 0000:04:00.0: irq 45 for MSI/MSI-X [  781.454634] IPv6:

> ADDRCONF(NETDEV_UP): p33p1: link is not ready [  782.893744] ATL1E

> 0000:04:00.0 p33p1: NIC Link is Up <100 Mbps Full Duplex> [  782.893797] IPv6:

> ADDRCONF(NETDEV_CHANGE): p33p1: link becomes ready [  783.194443]

> ATL1E 0000:04:00.0 p33p1: rx sequence number error (rx=0) (expect=1)

> [  783.204497] ATL1E 0000:04:00.0 p33p1: NIC Link is Up <100 Mbps Full Duplex>

> [  783.225336] ATL1E 0000:04:00.0 p33p1: rx sequence number error (rx=0)

> (expect=1) [  783.236636] ATL1E 0000:04:00.0 p33p1: NIC Link is Up <100 Mbps

> Full Duplex> [  783.462652] ATL1E 0000:04:00.0 p33p1: rx sequence number

> error (rx=0) (expect=1) [  783.472530] ATL1E 0000:04:00.0 p33p1: NIC Link is

> Up <100 Mbps Full Duplex> [  784.038976] ATL1E 0000:04:00.0 p33p1: rx

> sequence number error (rx=0) (expect=1) [  784.047354] ATL1E 0000:04:00.0

> p33p1: NIC Link is Up <100 Mbps Full Duplex> ...
Huang, Xiong March 28, 2013, 3:52 a.m. UTC | #4 
Sorry, my fault, your code is right.

> -----Original Message-----

> From: Hannes Frederic Sowa [mailto:hannes@stressinduktion.org]

> Sent: Thursday, March 28, 2013 11:30 AM

> To: Huang, Xiong

> Cc: Sven Hartge; netdev@vger.kernel.org

> Subject: Re: Atheros Communications Inc. AR8121/AR8113/AR8114 Gigabit or

> Fast Ethernet (rev b0) 1.0.0.7 md5/sha1 corrupted using NFS and samba

> (updated) Version 2

> 

> On Thu, Mar 28, 2013 at 03:25:29AM +0000, Huang, Xiong wrote:

> > Please don't revise following code:

> >

> > 			rx_page->read_offset +=

> > 				(((u32)((prrs->word1 >>

> RRS_PKT_SIZE_SHIFT) &

> > 				RRS_PKT_SIZE_MASK) +

> > 				sizeof(struct atl1e_recv_ret_status) + 31) &

> > 						0xFFFFFFE0);

> >

> > Here 32bytes alignment is no matter with Page alignment.

> 

> If I revert this change I get sequence errors:

> 

> [  781.449864] ATL1E 0000:04:00.0: irq 45 for MSI/MSI-X [  781.454634] IPv6:

> ADDRCONF(NETDEV_UP): p33p1: link is not ready [  782.893744] ATL1E

> 0000:04:00.0 p33p1: NIC Link is Up <100 Mbps Full Duplex> [  782.893797] IPv6:

> ADDRCONF(NETDEV_CHANGE): p33p1: link becomes ready [  783.194443]

> ATL1E 0000:04:00.0 p33p1: rx sequence number error (rx=0) (expect=1)

> [  783.204497] ATL1E 0000:04:00.0 p33p1: NIC Link is Up <100 Mbps Full Duplex>

> [  783.225336] ATL1E 0000:04:00.0 p33p1: rx sequence number error (rx=0)

> (expect=1) [  783.236636] ATL1E 0000:04:00.0 p33p1: NIC Link is Up <100 Mbps

> Full Duplex> [  783.462652] ATL1E 0000:04:00.0 p33p1: rx sequence number

> error (rx=0) (expect=1) [  783.472530] ATL1E 0000:04:00.0 p33p1: NIC Link is

> Up <100 Mbps Full Duplex> [  784.038976] ATL1E 0000:04:00.0 p33p1: rx

> sequence number error (rx=0) (expect=1) [  784.047354] ATL1E 0000:04:00.0

> p33p1: NIC Link is Up <100 Mbps Full Duplex> ...
Hannes Frederic Sowa March 29, 2013, 3:05 a.m. UTC | #5 
On Thu, Mar 28, 2013 at 03:39:33AM +0000, Huang, Xiong wrote:
> Dump every packet length & read_offset value,  compare them.

I think we have it! :)

If we disable msi as it is done in the windows driver, everything works as
expected. Do you have any clue why this could be the case? Otherwise I would
send a patch to disable msi by default.

Thanks for your help,

  Hannes

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Huang, Xiong March 29, 2013, 3:39 a.m. UTC | #6 
> 

> On Thu, Mar 28, 2013 at 03:39:33AM +0000, Huang, Xiong wrote:

> > Dump every packet length & read_offset value,  compare them.

> 

> I think we have it! :)

> 

> If we disable msi as it is done in the windows driver, everything works as

> expected. Do you have any clue why this could be the case? Otherwise I

> would send a patch to disable msi by default.

> 


Windows doesn't deliberately disable MSI, it just simply use INTx style since this chip hasn't MSIX capability :)
I think it's a safe solution to disable its MSI.

Thank you for all the test !

-Xiong
diff mbox

Patch

diff --git a/drivers/net/ethernet/atheros/atl1e/atl1e_main.c b/drivers/net/ethernet/atheros/atl1e/atl1e_main.c
index e1f1b2a..5264810 100644
--- a/drivers/net/ethernet/atheros/atl1e/atl1e_main.c
+++ b/drivers/net/ethernet/atheros/atl1e/atl1e_main.c
@@ -697,7 +697,7 @@  static void atl1e_cal_ring_size(struct atl1e_adapter *adapter, u32 *ring_size)
 		     sizeof(struct atl1e_tpd_desc) + 7
 			/* tx ring, qword align */
 		     + adapter->rx_ring.real_page_size * AT_PAGE_NUM_PER_QUEUE *
-			adapter->num_rx_queues + 31
+			adapter->num_rx_queues + 255
 			/* rx ring,  32 bytes align */
 		     + (1 + AT_PAGE_NUM_PER_QUEUE * adapter->num_rx_queues) *
 			sizeof(u32) + 3));
@@ -714,7 +714,7 @@  static void atl1e_init_ring_resources(struct atl1e_adapter *adapter)
 				 + adapter->hw.max_frame_size
 				 + ETH_HLEN + VLAN_HLEN
 				 + ETH_FCS_LEN;
-	rx_ring->real_page_size = roundup(rx_ring->real_page_size, 32);
+	rx_ring->real_page_size = roundup(rx_ring->real_page_size, 256);
 	atl1e_cal_ring_size(adapter, &adapter->ring_size);
 
 	adapter->ring_vir_addr = NULL;
@@ -825,7 +825,7 @@  static int atl1e_setup_ring_resources(struct atl1e_adapter *adapter)
 
 	/* Init RXF-Pages */
 	offset += (sizeof(struct atl1e_tpd_desc) * tx_ring->count);
-	offset = roundup(offset, 32);
+	offset = roundup(offset, 256);
 
 	for (i = 0; i < adapter->num_rx_queues; i++) {
 		for (j = 0; j < AT_PAGE_NUM_PER_QUEUE; j++) {
@@ -1003,7 +1003,7 @@  static inline void atl1e_configure_rx(struct atl1e_adapter *adapter)
 		rxq_ctrl_data |=
 			(RXQ_CTRL_HASH_ENABLE | RXQ_CTRL_RSS_MODE_MQUESINT);
 
-	rxq_ctrl_data |= RXQ_CTRL_IPV6_XSUM_VERIFY_EN | RXQ_CTRL_PBA_ALIGN_32 |
+	rxq_ctrl_data |= RXQ_CTRL_IPV6_XSUM_VERIFY_EN | RXQ_CTRL_PBA_ALIGN_256 |
 			 RXQ_CTRL_CUT_THRU_EN | RXQ_CTRL_EN;
 
 	AT_WRITE_REG(hw, REG_RXQ_CTRL, rxq_ctrl_data);
@@ -1444,8 +1444,8 @@  skip_pkt:
 			rx_page->read_offset +=
 				(((u32)((prrs->word1 >> RRS_PKT_SIZE_SHIFT) &
 				RRS_PKT_SIZE_MASK) +
-				sizeof(struct atl1e_recv_ret_status) + 31) &
-						0xFFFFFFE0);
+				sizeof(struct atl1e_recv_ret_status) + 255) &
+						0xFFFFFF00);
 
 			if (rx_page->read_offset >= rx_ring->page_size) {
 				/* mark this page clean */