Re: STMMAC driver: NFS Problem on 2.6.37
From: Shiraz Hashim <hidden>
Date: 2011-01-28 12:43:31
Hello Chuck, On Wed, Jan 26, 2011 at 02:04:21AM +0800, Chuck Lever wrote:
See analysis in line. On Jan 25, 2011, at 6:56 AM, deepaksi wrote:
[...]
quoted
We have made following observations 1. It seems that the time taken by phy auto negotiation process is long and as soon as the link gets up rpc ping request is getting timed out and we receive "Unable to reach ICMP" error. The time out error is same even if you do not connect a network cable and do a nfs boot. 2. We tried to modify the rate at which the work queue is scheduled in the phy framework. instead of scheduling every HZ ( 1 sec), we modified it to HZ/10. We did not received the error. This probably reduced the margin of the phy framework informing the kernel that the link is up. 3. We tried to use another network card and did a nfs boot. The only relevant difference we could find was the time of auto negotiation.Can you post a similar debugging dump of a root mount that succeeds using a different network card?
Following is the NFS boot log with a PCIe based e1000e nic card. .... .... [ 1.570000] e1000: Intel(R) PRO/1000 Network Driver - version 7.3.21-k8-NAPI [ 1.580000] e1000: Copyright (c) 1999-2006 Intel Corporation. [ 1.590000] e1000e: Intel(R) PRO/1000 Network Driver - 1.2.7-k2 [ 1.590000] e1000e: Copyright (c) 1999 - 2010 Intel Corporation. [ 1.600000] e1000e 0000:01:00.0: Disabling ASPM L1 [ 1.600000] PCI: enabling device 0000:01:00.0 (0140 -> 0142) [ 1.610000] e1000e 0000:01:00.0: (unregistered net_device): Failed to initialize MSI interrupts. Falling back to legacy interrupts. [ 1.850000] e1000e 0000:01:00.0: eth0: (PCI Express:2.5GB/s:Width x1) 00:15:17:ec:02:ff [ 1.860000] e1000e 0000:01:00.0: eth0: Intel(R) PRO/1000 Network Connection [ 1.870000] e1000e 0000:01:00.0: eth0: MAC: 1, PHY: 4, PBA No: d50861-004 [ 1.870000] Intel(R) Gigabit Ethernet Network Driver - version 2.1.0-k2 [ 1.880000] Copyright (c) 2007-2009 Intel Corporation. [ 1.880000] Intel(R) Virtual Function Network Driver - version 1.0.0-k0 [ 1.890000] Copyright (c) 2009 Intel Corporation. [ 1.900000] CAN device driver interface [ 1.900000] STMMAC driver: [ 1.900000] platform registration... [ 1.910000] done! [ 1.910000] DWMAC1000 - user ID: 0x10, Synopsys ID: 0x35 [ 1.920000] Enhanced descriptor structure [ 1.920000] eth1 - (dev. name: stmmaceth - id: 0, IRQ #65 [ 1.920000] IO base addr: 0xd00f0000) [ 1.940000] STMMAC MII Bus: probed [ 1.940000] eth1: PHY ID 20005c7a at 5 IRQ -1 (0:05) active [ 1.950000] ehci_hcd: USB 2.0 'Enhanced' Host Controller (EHCI) Driver [ 1.950000] spear-ehci spear-ehci.0: SPEAr EHCI [ 1.960000] spear-ehci spear-ehci.0: new USB bus registered, assigned bus number 1 [ 2.020000] spear-ehci spear-ehci.0: irq 96, io mem 0xe4800000 [ 2.040000] spear-ehci spear-ehci.0: USB 0.0 started, EHCI 1.00 [ 2.040000] hub 1-0:1.0: USB hub found [ 2.050000] hub 1-0:1.0: 1 port detected [ 2.050000] spear-ehci spear-ehci.1: SPEAr EHCI [ 2.060000] spear-ehci spear-ehci.1: new USB bus registered, assigned bus number 2 [ 2.120000] spear-ehci spear-ehci.1: irq 98, io mem 0xe5800000 [ 2.140000] spear-ehci spear-ehci.1: USB 0.0 started, EHCI 1.00
quoted
Are there some changes in the kernel framework w.r.t rpc ping time out ? This problem was not there in previous kernels.There have been changes in the RPC socket code around how it manages recovery from failed attempts to connect. We also have new logic now in the RPC client that causes RPC ping to fail immediately if a host can't be reached. Thanks for your efforts so far. It would be helpful if you could bisect to determine which commit(s) introduced this RPC client behavior (or any related changes to your network driver behavior).
We would do so in coming days. By then can you please guide us if you have something in mind. -- regards Shiraz -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org More majordomo info at http://vger.kernel.org/majordomo-info.html