Re: CURRENT: em0 NIC freezes under heavy I/O on net

From: Matthew Macy <mmacy_at_nextbsd.org>
Date: Wed, 11 Jan 2017 03:08:57 -0800
        

        
            Sorry, I meant to send that to the other thread. Was this after the iflib driver commit? If so it's odd that we haven't seen anything like this and I'll try to get a fix in ASAP.---- On Wed, 11 Jan 2017 03:06:19 -0800  Me<mmacy_at_nextbsd.org> wrote ----It looks like I have the wrong msix bar value for your NIC. Will fix in the next day or so.-M---- On Wed, 11 Jan 2017 00:27:30 -0800  O. Hartmann<ohartmann_at_walstatt.org> wrote ----Running recent CURRENT (FreeBSD 12.0-CURRENT #5 r311919: Wed Jan 11 08:24:28 CET 2017 amd64), the system freezes when doing a rsync over automounted (autofs) NFSv4 filesystem, mounted from another CURRENT server (same revision, but with BCM NICs).  The host in question is a Fujitsu Celsius M740 equipted with an Intel NIC:  [...] em0: <Intel(R) PRO/1000 Network Connection> port 0xf020-0xf03f mem 0xfb300000-0xfb31ffff,0xfb339000-0xfb339fff at device 25.0 numa-domain 0 on pci1 em0: attach_pre capping queues at 1 em0: using 1024 tx descriptors and 1024 rx descriptors em0: msix_init qsets capped at 1 em0: Unable to map MSIX table  em0: Using an MSI interrupt em0: allocated for 1 tx_queues em0: allocated for 1 rx_queues em0: netmap queues/slots: TX 1/1024, RX 1/1024 [...]  The pciconf output reveals:  em0_at_pci0:0:25:0:        class=0x020000 card=0x11ed1734 chip=0x153a8086 rev=0x05 hdr=0x00 vendor     = 'Intel Corporation'     device     = 'Ethernet Connection I217-LM'     class      = network     subclass   = ethernet     bar   [10] = type Memory, range 32, base 0xfb300000, size 131072, enabled     bar   [14] = type Memory, range 32, base 0xfb339000, size 4096, enabled     bar   [18] = type I/O Port, range 32, base 0xf020, size 32, enabled     cap 01[c8] = powerspec 2  supports D0 D3  current D0     cap 05[d0] = MSI supports 1 message, 64 bit enabled with 1 message     cap 13[e0] = PCI Advanced Features: FLR TP  I have a customized kernel. The NIC has revealed itself all the time as an "emX" device (never as igbX). The kernel contains device netmap (if relevevant).  The phenomenon:  Syncing a poudriere repository between to remote hosts, I use rsync on a NGSv4 exported filesystem, mounted via AUTOFS. So far, this work two days ago perfectly. Since yesterday, syncing brings down the network connection - the connection is simply dead. Terminating the rsync, bringing em0 down and up again doesn't help much, for short moments, the connection is established, but dies within seconds. Restarting via "service netif restart" all network services have the same effect: after the desaster, it is impossible for me to bring back the NIC/connection to normal, I have to reboot. The same happens when having heavy network load, but it takes a time and even rsync isn't "deadly" within the same timeframe - it takes sometimes a couple of seconds, another takes only one or two seconds to make the connection die.   I checked with dd'ing a large file over that connection, it takes several seconds then to make the connection freezing (so, someone could reproduce iy not ncessarily using rsync).  Kind regards,  oh _______________________________________________ freebsd-current_at_freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscribe_at_freebsd.org" 
        
        

    
    
Received on Wed Jan 11 2017 - 10:09:00 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:41:09 UTC