Re: where is the ethernet device number determined?
On 05/12/2016 02:19 PM, Cong Wang wrote: On Thu, May 12, 2016 at 1:05 PM, Chris Friesen <cbf...@mail.usask.ca> wrote: I hope this is a simple question...with legacy naming ethernet devices are named ethX. Where is that X determined? I've been looking in alloc_netdev_mqs() and friends, but haven't found it yet. __dev_alloc_name() Much appreciated. Chris
where is the ethernet device number determined?
Hi, I hope this is a simple question...with legacy naming ethernet devices are named ethX. Where is that X determined? I've been looking in alloc_netdev_mqs() and friends, but haven't found it yet. Thanks, Chris
[PATCH v4] route: do not cache fib route info on local routes with oif
For local routes that require a particular output interface we do not want to cache the result. Caching the result causes incorrect behaviour when there are multiple source addresses on the interface. The end result being that if the intended recipient is waiting on that interface for the packet he won't receive it because it will be delivered on the loopback interface and the IP_PKTINFO ipi_ifindex will be set to the loopback interface as well. This can be tested by running a program such as "dhcp_release" which attempts to inject a packet on a particular interface so that it is received by another program on the same board. The receiving process should see an IP_PKTINFO ipi_ifndex value of the source interface (e.g., eth1) instead of the loopback interface (e.g., lo). The packet will still appear on the loopback interface in tcpdump but the important aspect is that the CMSG info is correct. Sample dhcp_release command line: dhcp_release eth1 192.168.204.222 02:11:33:22:44:66 Signed-off-by: Allain Legacy <allain.leg...@windriver.com> Signed off-by: Chris Friesen <chris.frie...@windriver.com> --- net/ipv4/route.c | 12 1 file changed, 12 insertions(+) diff --git a/net/ipv4/route.c b/net/ipv4/route.c index 02c6229..b050cf9 100644 --- a/net/ipv4/route.c +++ b/net/ipv4/route.c @@ -2045,6 +2045,18 @@ static struct rtable *__mkroute_output(const struct fib_result *res, */ if (fi && res->prefixlen < 4) fi = NULL; + } else if ((type == RTN_LOCAL) && (orig_oif != 0) && + (orig_oif != dev_out->ifindex)) { + /* For local routes that require a particular output interface +* we do not want to cache the result. Caching the result +* causes incorrect behaviour when there are multiple source +* addresses on the interface, the end result being that if the +* intended recipient is waiting on that interface for the +* packet he won't receive it because it will be delivered on +* the loopback interface and the IP_PKTINFO ipi_ifindex will +* be set to the loopback interface as well. +*/ + fi = NULL; } fnhe = NULL;
Re: [PATCH v2] route: do not cache fib route info on local routes with oif
On 04/08/2016 01:14 PM, Julian Anastasov wrote: Your patch is corrupted. I was in the same trap some time ago but with different client: From Documentation/email-clients.txt: Don't send patches with "format=flowed". This can cause unexpected and unwanted line breaks. Anyways, the change looks good to me and I'll add my Reviewed-by tag the next time. Doh...forgot to turn off word wrapping. New patch coming. Chris
[PATCH v3] route: do not cache fib route info on local routes with oif
For local routes that require a particular output interface we do not want to cache the result. Caching the result causes incorrect behaviour when there are multiple source addresses on the interface. The end result being that if the intended recipient is waiting on that interface for the packet he won't receive it because it will be delivered on the loopback interface and the IP_PKTINFO ipi_ifindex will be set to the loopback interface as well. This can be tested by running a program such as "dhcp_release" which attempts to inject a packet on a particular interface so that it is received by another program on the same board. The receiving process should see an IP_PKTINFO ipi_ifndex value of the source interface (e.g., eth1) instead of the loopback interface (e.g., lo). The packet will still appear on the loopback interface in tcpdump but the important aspect is that the CMSG info is correct. Sample dhcp_release command line: dhcp_release eth1 192.168.204.222 02:11:33:22:44:66 Signed-off-by: Allain Legacy <allain.leg...@windriver.com> Signed off-by: Chris Friesen <chris.frie...@windriver.com> --- net/ipv4/route.c | 12 1 file changed, 12 insertions(+) diff --git a/net/ipv4/route.c b/net/ipv4/route.c index 02c6229..437a377 100644 --- a/net/ipv4/route.c +++ b/net/ipv4/route.c @@ -2045,6 +2045,18 @@ static struct rtable *__mkroute_output(const struct fib_result *res, */ if (fi && res->prefixlen < 4) fi = NULL; + } else if ((type == RTN_LOCAL) && (orig_oif != 0) && + (orig_oif != dev_out->ifindex)) { + /* For local routes that require a particular output interface + * we do not want to cache the result. Caching the result + * causes incorrect behaviour when there are multiple source + * addresses on the interface, the end result being that if the + * intended recipient is waiting on that interface for the + * packet he won't receive it because it will be delivered on + * the loopback interface and the IP_PKTINFO ipi_ifindex will + * be set to the loopback interface as well. +*/ + fi = NULL; } fnhe = NULL;
[PATCH v2] route: do not cache fib route info on local routes with oif
For local routes that require a particular output interface we do not want to cache the result. Caching the result causes incorrect behaviour when there are multiple source addresses on the interface. The end result being that if the intended recipient is waiting on that interface for the packet he won't receive it because it will be delivered on the loopback interface and the IP_PKTINFO ipi_ifindex will be set to the loopback interface as well. This can be tested by running a program such as "dhcp_release" which attempts to inject a packet on a particular interface so that it is received by another program on the same board. The receiving process should see an IP_PKTINFO ipi_ifndex value of the source interface (e.g., eth1) instead of the loopback interface (e.g., lo). The packet will still appear on the loopback interface in tcpdump but the important aspect is that the CMSG info is correct. Sample dhcp_release command line: dhcp_release eth1 192.168.204.222 02:11:33:22:44:66 Signed-off-by: Allain Legacy <allain.leg...@windriver.com> Signed off-by: Chris Friesen <chris.frie...@windriver.com> --- net/ipv4/route.c | 12 1 file changed, 12 insertions(+) diff --git a/net/ipv4/route.c b/net/ipv4/route.c index 02c6229..437a377 100644 --- a/net/ipv4/route.c +++ b/net/ipv4/route.c @@ -2045,6 +2045,18 @@ static struct rtable *__mkroute_output(const struct fib_result *res, */ if (fi && res->prefixlen < 4) fi = NULL; + } else if ((type == RTN_LOCAL) && (orig_oif != 0) && + (orig_oif != dev_out->ifindex)) { + /* For local routes that require a particular output interface +* we do not want to cache the result. Caching the result +* causes incorrect behaviour when there are multiple source +* addresses on the interface, the end result being that if the +* intended recipient is waiting on that interface for the +* packet he won't receive it because it will be delivered on +* the loopback interface and the IP_PKTINFO ipi_ifindex will +* be set to the loopback interface as well. +*/ + fi = NULL; } fnhe = NULL;
Re: [RFC PATCH] possible bug in handling of ipv4 route caching
On 04/07/2016 03:20 PM, Julian Anastasov wrote: On Thu, 7 Apr 2016, Chris Friesen wrote: Hi, We think we may have found a bug in the handling of ipv4 route caching, and are curious what you think. For local routes that require a particular output interface we do not want to cache the result. Caching the result causes incorrect behaviour when there are multiple source addresses on the interface. The end result being that if the intended recipient is waiting on that interface for the packet he won't receive it because it will be delivered on the loopback interface and the IP_PKTINFO ipi_ifindex will be set to the loopback interface as well. diff --git a/net/ipv4/route.c b/net/ipv4/route.c index 02c6229..e965d4b 100644 --- a/net/ipv4/route.c +++ b/net/ipv4/route.c @@ -2045,6 +2045,17 @@ static struct rtable *__mkroute_output(const struct fib_result *res, */ if (fi && res->prefixlen < 4) fi = NULL; + } else if ((type == RTN_LOCAL) && (orig_oif != 0)) { So, we can be more specific. Can this work?: } else if ((type == RTN_LOCAL) && (orig_oif != 0) && (orig_oif != dev_out->ifindex)) { I.e. we should allow to cache orig_oif=LOOPBACK_IFINDEX but eth1 should not be cached. Yes, we think that will work. New patch to follow. Chris
[RFC PATCH] possible bug in handling of ipv4 route caching
Hi, We think we may have found a bug in the handling of ipv4 route caching, and are curious what you think. For local routes that require a particular output interface we do not want to cache the result. Caching the result causes incorrect behaviour when there are multiple source addresses on the interface. The end result being that if the intended recipient is waiting on that interface for the packet he won't receive it because it will be delivered on the loopback interface and the IP_PKTINFO ipi_ifindex will be set to the loopback interface as well. This can be tested by running a program such as "dhcp_release" which attempts to inject a packet on a particular interface so that it is received by another program on the same board. The receiving process should see an IP_PKTINFO ipi_ifndex value of the source interface (e.g., eth1) instead of the loopback interface (e.g., lo). The packet will still appear on the loopback interface in tcpdump but the important aspect is that the CMSG info is correct. For what it's worth, here's a patch that we've applied locally to deal with the issue. Chris Signed-off-by: Allain Legacy <allain.leg...@windriver.com> Signed-off-by: Chris Friesen <chris.frie...@windriver.com> diff --git a/net/ipv4/route.c b/net/ipv4/route.c index 02c6229..e965d4b 100644 --- a/net/ipv4/route.c +++ b/net/ipv4/route.c @@ -2045,6 +2045,17 @@ static struct rtable *__mkroute_output(const struct fib_result *res, */ if (fi && res->prefixlen < 4) fi = NULL; + } else if ((type == RTN_LOCAL) && (orig_oif != 0)) { + /* For local routes that require a particular output interface + * we do not want to cache the result. Caching the result + * causes incorrect behaviour when there are multiple source + * addresses on the interface, the end result being that if the + * intended recipient is waiting on that interface for the + * packet he won't receive it because it will be delivered on + * the loopback interface and the IP_PKTINFO ipi_ifindex will + * be set to the loopback interface as well. +*/ + fi = NULL; } fnhe = NULL;
Re: questions on NAPI processing latency and dropped network packets
I've done some further digging, and it appears that one of the problems we may be facing is very high instantaneous traffic rates. Instrumentation showed up to 222K packets/sec for short periods (at least 1.1 ms, possibly longer), although the long-term average is down around 14-16K packets/sec. If I bump the rx ring size up to 4096, we can handle all the packets and we still have 44% idle on cpu0 and 27% idle on cpu1. Is there anything else we can do to minimize the latency of network packet processing and avoid having to crank the rx ring size up so high? Thanks, Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: questions on NAPI processing latency and dropped network packets
Ben Greear wrote: Chris Friesen wrote: Is there anything else we can do to minimize the latency of network packet processing and avoid having to crank the rx ring size up so high? Why is it such a big deal to crank up the rx queue length? Seems like a perfectly normal way to handle bursts like this... It means that the latency for handling those packets is higher than it could be. Draining 4096 packets from the queue will take a while. Ideally we'd like to bring the latency down as much as possible, and then bump up the rx queue length to handle the rest. Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: questions on NAPI processing latency and dropped network packets
Eric Dumazet wrote: Chris Friesen a écrit : I've done some further digging, and it appears that one of the problems we may be facing is very high instantaneous traffic rates. Instrumentation showed up to 222K packets/sec for short periods (at least 1.1 ms, possibly longer), although the long-term average is down around 14-16K packets/sec. Instrumentation done where exactly ? I added some code to e1000_clean_rx_irq() to track rx_fifo drops, total packets received, and an accurate timestamp. If rx_fifo errors changed, it would dump the information. Is there anything else we can do to minimize the latency of network packet processing and avoid having to crank the rx ring size up so high? You have some tasks that disable softirqs too long. Sometimes, bumping RX ring size is OK (but you will still have delays), sometimes it is not an option, since 4096 is the limit on current hardware. I added some instrumentation to take timestamps in __do_softirq() as well. Based on these timestamps, I can see the following code sequence: 2374604616 usec, start processing softirqs in __do_softirq() 2374610337 usec, log values in e1000_clean_rx_irq() 2374611411 usec, log values in e1000_clean_rx_irq() In between the successive calls to e1000_clean_rx_irq() the rx_fifo counts went up. Does anyone have any patchsets to track down what softirqs are taking a long time, and/or who's disabling softirqs? Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: questions on NAPI processing latency and dropped network packets
Jarek Poplawski wrote: IMHO, checking this with a current stable, which probably you are going to do some day, anyway, should be 100% acceptable: giving some input to netdev, while still working for yourself. While I would love to do this, it's not that simple. Some of our hardware is not supported on mainline, so we need per-kernel version patches to even bring up the blade. The blades netboot via a jumbo-frame network, so kernel extensions are needed to handle setting the MTU before mounting the rootfs. The blade in question uses CKRM which doesn't exist for newer kernels so the problem may simply be hidden by scheduling differences. The userspace application uses other kernel features that are not in mainline (and likely some of them won't ever be in mainline--I know because I've tried). Basically, the number of changes required for our environment makes it difficult to just boot up the latest kernel. Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: questions on NAPI processing latency and dropped network packets
Radoslaw Szkodzinski (AstralStorm) wrote: On Tue, 15 Jan 2008 08:47:07 -0600 Chris Friesen [EMAIL PROTECTED] wrote: Some of our hardware is not supported on mainline, so we need per-kernel version patches to even bring up the blade. The blades netboot via a jumbo-frame network, so kernel extensions are needed to handle setting the MTU before mounting the rootfs. Why? Can't you use a small initramfs to set it up? Sure, if we changed our build system, engineered a suitable small userspace, etc. At this point it's easier to maintain a small patch to the kernel dhcp parsing code so that we can specify mtu. The blade in question uses CKRM which doesn't exist for newer kernels so the problem may simply be hidden by scheduling differences. Current spiritual successor is Ingo's realtime patchset I guess. I think the group scheduling stuff for CFS is closer, but there are design and API differences that would require us to rework our system. The userspace application uses other kernel features that are not in mainline (and likely some of them won't ever be in mainline--I know because I've tried). What would these be? Some proc or sysfs files that were removed or renamed? Maybe they can be worked around w/o changing the application at all, or very minor changes. No, more than proc/sysfs. Things like notification of process state change (think like SIGCHLD but sent to arbitrary processes), additional messaging protocol families, oom-killer protection, dirty page monitoring, backwards compatibility for dcbz on the ppc970, nested alternate signal stacks, and many others. Between our kernel vendor's patches and our own, there are something like 600 patches applied to the mainline kernel. Also, be sure to check if there are pauses with other CPU hogs. With the sctp hash patch applied we're now sitting with 45% cpu free on one cpu, and about 25% free on the other, and we're still seeing periodic bursts of rx fifo loss. It's wierd. Still trying to figure out what's going on. Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: questions on NAPI processing latency and dropped network packets
Ray Lee wrote: On Jan 10, 2008 9:24 AM, Chris Friesen [EMAIL PROTECTED] wrote: After a recent userspace app change, we've started seeing packets being dropped by the ethernet hardware (e1000, NAPI is enabled). The error/dropped/fifo counts are going up in ethtool: Can you reproduce it with a simple userspace cpu hog? (Two, really, one per cpu.) Can you reproduce it with the newer e1000? Hmm...good questions and I haven't checked either. The first one is relatively straightforward. The second is a bit trickier...last time I tried the latest e1000 driver the card wouldn't boot (we use netboot). Can you reproduce it with git head? Unfortunately, I don't think I'll be able to try this. We require kernel mods for our userspace to run, and I doubt I'd be able to get the time to port all the changes forward to git head. If the answer to the first one is yes, the last no, then bisect until you get a kernel that doesn't show the problem. Backport the fix, unless the fix happens to be CFS. However, I suspect that your userpace app is just starving the system from time to time. It's conceivable that userspace is starving the kernel, but we have do about 45% idle on one cpu, and 7-10% idle on the other. We also have an odd situation where on an initial test run after bootup we have 18-24% idle on cpu1, but resetting the test tool drops that to the 7-10% I mentioned above. Based on profiling and instrumentation it seems like the cost of sctp_endpoint_lookup_assoc() more than triples, which means that the amount of time that bottom halves are disabled in that function also triples. -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: questions on NAPI processing latency and dropped network packets
David Miller wrote: From: Chris Friesen [EMAIL PROTECTED] Date: Fri, 11 Jan 2008 08:59:26 -0600 I'd love to work on newer kernels, but we have a commitment to our customers to support multiple releases for a significant amount of time. And by asking here for people to dig into it for you, you are asking people for free help providing that support. I'm not asking people to spend significant amounts of time...more like if anyone has any ideas off the top of their heads. That's why there is such negative backlash to asking questions about such ancient kernel here, you're asking us to do your work, for free. I hadn't realized that you felt this strongly about asking questions regarding old kernels. How close to bleeding edge do we need to be for it to be considered acceptable to ask questions on netdev? Given that the embedded space tends to be perpetually stuck on older kernels (our current release is based on 2.6.14) do you have any suggestion on how we can obtain information that would be available on netdev if we were using newer kernels? Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: questions on NAPI processing latency and dropped network packets
Eric Dumazet wrote: Chris Friesen a écrit : Based on profiling and instrumentation it seems like the cost of sctp_endpoint_lookup_assoc() more than triples, which means that the amount of time that bottom halves are disabled in that function also triples. Any idea of the size of sctp hash size you have ? (your dmesg probably includes a message starting with SCTP: Hash tables configured... How many concurrent sctp sockets are handled ? Our lab is currently rebooting, but I'll try and get this once it's back up. Maybe sctp_assoc_hashfn() is too weak for your use, and some chains are *really* long. Based on the profiling information we're spending time in sctp_endpoint_lookup_assoc() which doesn't actually use hashes, so I can't see how the hash would be related. I'm pretty new to SCTP though, so I may be missing something. Here's the top results from readprofile, unfortunately these are aggregated across both cpus so they don't really show what's going on. The key thing is that sctp_endpoint_lookup_assoc() is the most expensive kernel routine on this entire system. 3147 .power4_idle 22.4786 1712 .native_idle 20.3810 1234 .sctp_endpoint_lookup_assoc2.1725 1212 ._spin_unlock_irqrestore 6.4468 778 .do_futex 0.3791 447 ._spin_unlock_irq 4.2981 313 .fget 1.7784 277 .fput 3.8472 275 .kfree 0.7473 234 .__kmalloc 0.5571 131 SystemCall_common 0.3411 130 .sctp_assoc_is_match 0.6373 123 .lock_sock 0.4155 119 .find_vma 0.6919 116 .kmem_cache_alloc 0.3580 111 .kmem_cache_free 0.3343 106 .skb_release_data 0.4907 102 .__copy_tofrom_user0.0724 100 .exit_elf_binfmt 1.9231 100 .do_select 0.0820 Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: questions on NAPI processing latency and dropped network packets
Eric Dumazet wrote: Chris Friesen a écrit : Based on the profiling information we're spending time in sctp_endpoint_lookup_assoc() which doesn't actually use hashes, so I can't see how the hash would be related. I'm pretty new to SCTP though, so I may be missing something. Well, it does use hashes :) hash = sctp_assoc_hashfn(ep-base.bind_addr.port, rport); head = sctp_assoc_hashtable[hash]; read_lock(head-lock); sctp_for_each_hentry(epb, node, head-chain) { /* maybe your machine is traversing here a *really* long chain */ } The latest released kernel doesn't have this code, it was only added in November. The SCTP maintainer just pointed me to the patch, and made some other suggestions as well. Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: questions on NAPI processing latency and dropped network packets
David Miller wrote: You have to be kidding, coming here for help with a nearly 4 year old kernel. I figured it couldn't hurt to ask...if I can't ask the original authors, who else is there? I'd love to work on newer kernels, but we have a commitment to our customers to support multiple releases for a significant amount of time. Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
questions on NAPI processing latency and dropped network packets
Hi all, I've got an issue that's popped up with a deployed system running 2.6.10. I'm looking for some help figuring out why incoming network packets aren't being processed fast enough. After a recent userspace app change, we've started seeing packets being dropped by the ethernet hardware (e1000, NAPI is enabled). The error/dropped/fifo counts are going up in ethtool: rx_packets: 32180834 rx_bytes: 5480756958 rx_errors: 862506 rx_dropped: 771345 rx_length_errors: 0 rx_over_errors: 0 rx_crc_errors: 0 rx_frame_errors: 0 rx_fifo_errors: 91161 rx_missed_errors: 91161 This link is receiving roughly 13K packets/sec, and we're dropping roughly 51 packets/sec due to fifo errors. Increasing the rx descriptor ring size from 256 up to around 3000 or so seems to make the problem stop, but it seems to me that this is just a workaround for the latency in processing the incoming packets. So, I'm looking for some suggestions on how to fix this or to figure out where the latency is coming from. Some additional information: 1) Interrupts are being processed on both cpus: [EMAIL PROTECTED]:/root cat /proc/interrupts CPU0 CPU1 30:17037564530785 U3-MPIC Level eth0 2) top shows a fair amount of time processing softirqs, but very little time in ksoftirqd (or is that a sampling artifact?). Tasks: 79 total, 1 running, 78 sleeping, 0 stopped, 0 zombie Cpu0: 23.6% us, 30.9% sy, 0.0% ni, 36.9% id, 0.0% wa, 0.3% hi, 8.3% si Cpu1: 30.4% us, 24.1% sy, 0.0% ni, 5.9% id, 0.0% wa, 0.7% hi, 38.9% si Mem: 4007812k total, 2199148k used, 1808664k free, 0k buffers Swap: 0k total, 0k used, 0k free, 219844k cached PID USER PR NI VIRT RES SHR S %CPU %MEMTIME+ COMMAND 5375 root 15 0 2682m 1.8g 6640 S 99.9 46.7 31:17.68 SigtranServices 7696 root 17 0 6952 3212 1192 S 7.3 0.1 0:15.75 schedmon.ppc210 7859 root 16 0 2688 1228 964 R 0.7 0.0 0:00.04 top 2956 root 8 -8 18940 7436 5776 S 0.3 0.2 0:01.35 blademtc 1 root 16 0 1660 620 532 S 0.0 0.0 0:30.62 init 2 root RT 0 000 S 0.0 0.0 0:00.01 migration/0 3 root 15 0 000 S 0.0 0.0 0:00.55 ksoftirqd/0 4 root RT 0 000 S 0.0 0.0 0:00.01 migration/1 5 root 15 0 000 S 0.0 0.0 0:00.43 ksoftirqd/1 3) /proc/sys/net/core/netdev_max_backlog is set to the default of 300 So...anyone have any ideas/suggestions? Thanks, Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: questions on NAPI processing latency and dropped network packets
Kok, Auke wrote: You're using 2.6.10... you can always replace the e1000 module with the out-of-tree version from e1000.sf.net, this might help a bit - the version in the 2.6.10 kernel is very very old. Do you have any reason to believe this would improve things? It seems like the problem lies in the NAPI/softirq code rather than in the e1000 driver itself, no? it also appears that your app is eating up CPU time. perhaps setting the app to a nicer nice level might mitigate things a bit. If we're not handling the softirq work from ksoftirqd how would changing scheduler settings affect anything? Also turn off the in-kernel irq mitigation, it just causes cache misses and you really need the network irq to sit on a single cpu at most (if not all) the time to get the best performance. Use the userspace irqbalance daemon instead to achieve this. Using userspace irqbalance would be some effort to test and deploy properly. However, as a quick test I tried setting the irq affinity for this device and it didn't help. One thing that might be of interest is that it seems to be bursty rather than gradual. Here are some timestamps (in seconds) along with the number of overruns on eth0: 6552.15 overruns:260097 6552.69 overruns:260097 6553.32 overruns:260097 6553.83 overruns:260097 6554.35 overruns:260097 6554.87 overruns:260097 6555.41 overruns:260097 6555.94 overruns:260097 6556.51 overruns:260097 6557.07 overruns:260282 6557.58 overruns:260282 6558.23 overruns:260282 Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: questions on NAPI processing latency and dropped network packets
James Chapman wrote: What's changed in your application? Any real-time threads in there? From the top output below, looks like SigtranServices is consuming all your CPU... There are two cpus, and SigtranServices is multithreaded with many threads. Most of these threads are affined to cpu0, a couple to cpu1. None of the threads are realtime. Top is showing 37% idle on cpu0, and 6% idle on cpu1, so not all the cpu is being consumed. However, I'm wondering if we're hitting bursty bits and we're just running out of time. I'm going to try a system with MAX_SOFTIRQ_RESTART bumped up a bit, and also enable profiling. Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: ip neigh show not showing arp cache entries?
Herbert Xu wrote: Chris Friesen [EMAIL PROTECTED] wrote: However, if I specifically try to print out one of the missing entries, it shows up: [EMAIL PROTECTED]:/root /tmp/ip neigh show 192.168.24.81 192.168.24.81 dev bond2 lladdr 00:01:af:14:e9:8a REACHABLE What about ip -4 neigh show Looks like that did it. Why does specifying the family make a difference? [EMAIL PROTECTED]:/root ip neigh show 10.41.18.1 dev eth6 lladdr 00:00:5e:00:01:01 STALE 172.24.137.0 dev bond0 lladdr 00:c0:8b:08:e4:88 REACHABLE 172.24.0.9 dev bond0 lladdr 00:07:e9:41:4b:b4 REACHABLE 10.41.18.101 dev eth6 lladdr 00:0e:0c:5e:95:bd REACHABLE 172.24.0.11 dev bond0 lladdr 00:03:cc:51:06:5e STALE 172.24.132.1 dev bond0 lladdr 00:01:af:14:e9:88 REACHABLE 172.24.0.15 dev bond0 lladdr 00:0e:0c:85:fd:d2 STALE 172.24.0.3 dev bond0 lladdr 00:01:af:14:c8:cc REACHABLE 172.24.0.5 dev bond0 lladdr 00:01:af:15:e0:6a STALE [EMAIL PROTECTED]:/root ip -4 neigh show 192.168.24.81 dev bond2 lladdr 00:01:af:14:e9:8a REACHABLE 172.24.132.2 dev bond0 FAILED 172.24.136.0 dev bond0 lladdr 00:c0:8b:07:b3:7e REACHABLE 10.41.18.1 dev eth6 lladdr 00:00:5e:00:01:01 STALE 172.24.137.0 dev bond0 lladdr 00:c0:8b:08:e4:88 REACHABLE 172.24.0.9 dev bond0 lladdr 00:07:e9:41:4b:b4 REACHABLE 10.41.18.101 dev eth6 lladdr 00:0e:0c:5e:95:bd REACHABLE 172.24.0.11 dev bond0 lladdr 00:03:cc:51:06:5e STALE 172.24.132.1 dev bond0 lladdr 00:01:af:14:e9:88 REACHABLE 172.24.0.15 dev bond0 lladdr 00:0e:0c:85:fd:d2 STALE 172.24.0.3 dev bond0 lladdr 00:01:af:14:c8:cc REACHABLE 172.24.0.5 dev bond0 lladdr 00:01:af:15:e0:6a STALE Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: ip neigh show not showing arp cache entries?
YOSHIFUJI Hideaki / 吉藤英明 wrote: In article [EMAIL PROTECTED] (at Wed, 12 Dec 2007 15:57:08 -0600), Chris Friesen [EMAIL PROTECTED] says: You may try other versions of this command http://devresources.linux-foundation.org/dev/iproute2/download/ They appear to be numbered by kernel version, and the above version is the most recent one for 2.6.14. Will more recent ones (for newer kernels) work with my kernel? It should work; if it doesn't, please make a report. Thanks. I downloaded iproute2-2.6.23 and built it for my kernel. I'm compiling for a different kernel than is actually running on the build system, so I had to add a line defining KERNEL_INCLUDE to the Makefile, and I had to add -I${KERNEL_INCLUDE} to the CFLAGS definition. Someone might want to do something about that... Anyways, the arp entry issue is still there. The arp command gives a bunch of entries: [EMAIL PROTECTED]:/root arp -n Address HWtype HWaddress Flags MaskIface 192.168.24.81ether 00:01:AF:14:E9:8A C bond2 172.24.132.2 (incomplete) bond0 172.24.136.0 ether 00:C0:8B:07:B3:7E C bond0 172.24.137.0 (incomplete) bond0 172.24.0.9 ether 00:07:E9:41:4B:B4 C bond0 10.41.18.101 ether 00:0E:0C:5E:95:BD C eth6 172.24.0.11 ether 00:03:CC:51:06:5E C bond0 172.24.132.1 ether 00:01:AF:14:E9:88 C bond0 172.24.0.15 ether 00:0E:0C:85:FD:D2 C bond0 172.24.0.3 ether 00:01:AF:14:C8:CC C bond0 172.24.0.5 ether 00:01:AF:15:E0:6A C bond0 The original ip command and the new one (/tmp/ip) both give the same results--some of the entries are missing. [EMAIL PROTECTED]:/root ip neigh show all 172.24.137.0 dev bond0 FAILED 172.24.0.9 dev bond0 lladdr 00:07:e9:41:4b:b4 REACHABLE 10.41.18.101 dev eth6 lladdr 00:0e:0c:5e:95:bd REACHABLE 172.24.0.11 dev bond0 lladdr 00:03:cc:51:06:5e STALE 172.24.132.1 dev bond0 lladdr 00:01:af:14:e9:88 REACHABLE 172.24.0.15 dev bond0 lladdr 00:0e:0c:85:fd:d2 STALE 172.24.0.3 dev bond0 lladdr 00:01:af:14:c8:cc REACHABLE 172.24.0.5 dev bond0 lladdr 00:01:af:15:e0:6a STALE [EMAIL PROTECTED]:/root /tmp/ip neigh show all 172.24.137.0 dev bond0 FAILED 172.24.0.9 dev bond0 lladdr 00:07:e9:41:4b:b4 REACHABLE 10.41.18.101 dev eth6 lladdr 00:0e:0c:5e:95:bd REACHABLE 172.24.0.11 dev bond0 lladdr 00:03:cc:51:06:5e STALE 172.24.132.1 dev bond0 lladdr 00:01:af:14:e9:88 REACHABLE 172.24.0.15 dev bond0 lladdr 00:0e:0c:85:fd:d2 STALE 172.24.0.3 dev bond0 lladdr 00:01:af:14:c8:cc REACHABLE 172.24.0.5 dev bond0 lladdr 00:01:af:15:e0:6a STALE However, if I specifically try to print out one of the missing entries, it shows up: [EMAIL PROTECTED]:/root /tmp/ip neigh show 192.168.24.81 192.168.24.81 dev bond2 lladdr 00:01:af:14:e9:8a REACHABLE Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: ip neigh show not showing arp cache entries?
Patrick McHardy wrote: From a kernel perspective there are only complete dumps, the filtering is done by iproute. So the fact that it shows them when querying specifically implies there is a bug in the iproute neighbour filter. Does it work if you omit all from the ip neigh show command? Omitting all gives identical results. It is still missing entries when compared with the output of arp. Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: ip neigh show not showing arp cache entries?
I retested it on an x86 machine and am seeing similar problems. First, arp gives the arp table as expected: [EMAIL PROTECTED]:/tftpboot/cnp/0-0-5-0/0-0-5-0 arp -n Address HWtype HWaddress Flags Mask Iface 172.24.0.9 ether 00:03:CC:51:06:5E C bond0 10.41.18.101 ether 00:0E:0C:5E:95:BD C eth6 172.24.137.0 ether 00:C0:8B:08:E4:88 C bond0 172.24.136.0 ether 00:C0:8B:07:B3:7E C bond0 10.41.18.1 ether 00:00:5E:00:01:01 C eth6 172.24.0.5 ether 00:01:AF:15:E0:6A C bond0 172.24.0.13 ether 00:0E:0C:85:FD:D2 C bond0 172.24.0.3 ether 00:01:AF:14:C8:CC C bond0 172.24.132.1 ether 00:01:AF:14:E9:88 C bond0 172.24.0.7 ether 00:07:E9:41:4B:B4 C bond0 192.168.24.81ether 00:01:AF:14:E9:8A C bond2 ip neigh show gives nothing, but if I search for specific addresses from the arp table listing they show up: [EMAIL PROTECTED]:/tftpboot/cnp/0-0-5-0/0-0-5-0 ip neigh show [EMAIL PROTECTED]:/tftpboot/cnp/0-0-5-0/0-0-5-0 ip neigh show 172.24.0.9 172.24.0.9 dev bond0 lladdr 00:03:cc:51:06:5e DELAY [EMAIL PROTECTED]:/tftpboot/cnp/0-0-5-0/0-0-5-0 ip neigh show 10.41.18.101 10.41.18.101 dev eth6 lladdr 00:0e:0c:5e:95:bd REACHABLE [EMAIL PROTECTED]:/tftpboot/cnp/0-0-5-0/0-0-5-0 ip neigh show 172.24.137.0 172.24.137.0 dev bond0 lladdr 00:c0:8b:08:e4:88 REACHABLE Is this expected behaviour? Thanks, Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: ip neigh show not showing arp cache entries?
Eric Dumazet wrote: Chris Friesen a écrit : Is this expected behaviour? Probably not... Still a 2.6.14 kernel ? Yep. Embedded hardware, so I'm unable to test with a more recent kernel. Could you send the result of : strace ip neigh show I've attached two strace runs, one of ip neigh show and one of ip neigh show 10.41.18.101. Chris [EMAIL PROTECTED]:/tftpboot/cnp/0-0-5-0/0-0-5-0 strace ip neigh show execve(/sbin/ip, [ip, neigh, show], [/* 14 vars */]) = 0 uname({sys=Linux, node=typhoon-base-unit0, ...}) = 0 brk(0) = 0x806b000 access(/etc/ld.so.preload, R_OK) = -1 ENOENT (No such file or directory) open(/etc/ld.so.cache, O_RDONLY) = 3 fstat64(3, {st_mode=S_IFREG|0644, st_size=58478, ...}) = 0 mmap2(NULL, 58478, PROT_READ, MAP_PRIVATE, 3, 0) = 0xb7f56000 close(3)= 0 open(/lib/libresolv.so.2, O_RDONLY) = 3 read(3, \177ELF\1\1\1\0\0\0\0\0\0\0\0\0\3\0\3\0\1\0\0\0\320\\0..., 512) = 512 fstat64(3, {st_mode=S_IFREG|0755, st_size=75541, ...}) = 0 mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xb7f55000 mmap2(NULL, 71816, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_DENYWRITE, 3, 0) = 0xb7f43000 mmap2(0xb7f51000, 8192, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0xe) = 0xb7f51000 mmap2(0xb7f53000, 6280, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0xb7f53000 close(3)= 0 open(/lib/libc.so.6, O_RDONLY)= 3 read(3, \177ELF\1\1\1\0\0\0\0\0\0\0\0\0\3\0\3\0\1\0\0\0\216O\1..., 512) = 512 fstat64(3, {st_mode=S_IFREG|0755, st_size=1407983, ...}) = 0 mmap2(NULL, 1146076, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_DENYWRITE, 3, 0) = 0xb7e2b000 mmap2(0xb7f39000, 32768, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0x10d) = 0xb7f39000 mmap2(0xb7f41000, 7388, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0xb7f41000 close(3)= 0 mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xb7e2a000 mprotect(0xb7f39000, 20480, PROT_READ) = 0 set_thread_area({entry_number:-1 - 6, base_addr:0xb7e2a6b0, limit:1048575, seg_32bit:1, contents:0, read_exec_only:0, limit_in_pages:1, seg_not_present:0, useable:1}) = 0 munmap(0xb7f56000, 58478) = 0 socket(PF_NETLINK, SOCK_RAW, 0) = 3 setsockopt(3, SOL_SOCKET, SO_SNDBUF, [32768], 4) = 0 setsockopt(3, SOL_SOCKET, SO_RCVBUF, [32768], 4) = 0 bind(3, {sa_family=AF_NETLINK, pid=0, groups=}, 12) = 0 getsockname(3, {sa_family=AF_NETLINK, pid=6150, groups=}, [12]) = 0 time(NULL) = 1197465643 sendto(3, \24\0\0\0\22\0\1\3,\340_G\0\0\0\0\0\0\0\0, 20, 0, {sa_family=AF_NETLINK, pid=0, groups=}, 12) = 20 recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=}, msg_iov(1)=[{\364\0\0\0\20\0\2\0,\340_G\6\30\0\0\0\0\1\0\1\0\0\0C\30..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3544 brk(0) = 0x806b000 brk(0x808c000) = 0x808c000 recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=}, msg_iov(1)=[{\24\0\0\0\3\0\2\0,\340_G\6\30\0\0\0\0\0\0\1\0\0\0C\30\0..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 20 sendto(3, \24\0\0\0\36\0\1\3-\340_G\0\0\0\0\0\0\0\0, 20, 0, {sa_family=AF_NETLINK, pid=0, groups=}, 12) = 20 recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=}, msg_iov(1)=[{[EMAIL PROTECTED]..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 264 recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=}, msg_iov(1)=[{[EMAIL PROTECTED]..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 20 exit_group(0) = ? Process 6150 detached [EMAIL PROTECTED]:/tftpboot/cnp/0-0-5-0/0-0-5-0 ip neigh show 10.41.18.101 10.41.18.101 dev eth6 lladdr 00:0e:0c:5e:95:bd REACHABLE [EMAIL PROTECTED]:/tftpboot/cnp/0-0-5-0/0-0-5-0 strace ip neigh show 10.41.18.101 execve(/sbin/ip, [ip, neigh, show, 10.41.18.101], [/* 14 vars */]) = 0 uname({sys=Linux, node=typhoon-base-unit0, ...}) = 0 brk(0) = 0x806b000 access(/etc/ld.so.preload, R_OK) = -1 ENOENT (No such file or directory) open(/etc/ld.so.cache, O_RDONLY) = 3 fstat64(3, {st_mode=S_IFREG|0644, st_size=58478, ...}) = 0 mmap2(NULL, 58478, PROT_READ, MAP_PRIVATE, 3, 0) = 0xb7fd5000 close(3)= 0 open(/lib/libresolv.so.2, O_RDONLY) = 3 read(3, \177ELF\1\1\1\0\0\0\0\0\0\0\0\0\3\0\3\0\1\0\0\0\320\\0..., 512) = 512 fstat64(3, {st_mode=S_IFREG|0755, st_size=75541, ...}) = 0 mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xb7fd4000 mmap2(NULL, 71816, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_DENYWRITE, 3, 0) = 0xb7fc2000 mmap2(0xb7fd, 8192, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0xe) = 0xb7fd mmap2(0xb7fd2000, 6280, PROT_READ|PROT_WRITE
Re: ip neigh show not showing arp cache entries?
Eric Dumazet wrote: And what is the version of ip command you have on this machine ? ip -V iproute2-ss051107 You may try other versions of this command http://devresources.linux-foundation.org/dev/iproute2/download/ They appear to be numbered by kernel version, and the above version is the most recent one for 2.6.14. Will more recent ones (for newer kernels) work with my kernel? Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
ip neigh show not showing arp cache entries?
I'm seeing some strange behaviour on a 2.6.14 ppc64 system. If I run ip neigh show it prints out nothing, but if I run arp then I see the other nodes on the local network. [EMAIL PROTECTED]:/root ip neigh show [EMAIL PROTECTED]:/root arp -n Address HWtype HWaddress Flags Mask Iface 172.24.132.0 ether 00:01:AF:14:E8:DA C bond0 172.24.132.1 (incomplete) bond0 172.24.136.0 ether 00:C0:8B:07:B3:7E C bond0 172.24.132.4 ether 00:01:AF:14:E8:DA C bond0 172.24.132.2 ether 00:01:AF:14:E8:DA C bond0 Any ideas what's going on here? Thanks, Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: ip neigh show not showing arp cache entries?
Chris Friesen wrote: I'm seeing some strange behaviour on a 2.6.14 ppc64 system. If I run ip neigh show it prints out nothing, but if I run arp then I see the other nodes on the local network. [EMAIL PROTECTED]:/root ip neigh show [EMAIL PROTECTED]:/root arp -n Address HWtype HWaddress Flags MaskIface 172.24.132.0 ether 00:01:AF:14:E8:DA Cbond0 172.24.132.1 (incomplete)bond0 172.24.136.0 ether 00:C0:8B:07:B3:7E Cbond0 172.24.132.4 ether 00:01:AF:14:E8:DA Cbond0 172.24.132.2 ether 00:01:AF:14:E8:DA Cbond0 Any ideas what's going on here? I've got some further information. If I look for a specific address, it seems to work: [EMAIL PROTECTED]:/root ip neigh show 172.24.136.0 172.24.136.0 dev bond0 lladdr 00:c0:8b:07:b3:7e REACHABLE In the above scenario, the arp cache lists the device as reachable via bond0. If I search the arp cache to see whether the address is reachable from one of bond0's slave devices, should it come back positive or negative? Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
when using arp monitoring with bonding, why use broadcast arps?
We have a network with a number of nodes using bonding with arp monitoring. The arp interval is set to 100ms. Unfortunately, the bonding code sends the arp packets to the hardware broadcast address, which means that the number of these arp packets seen by each node goes up with the number of nodes on the network. One of the nodes has a fairly low-powered cpu and handles most things in microengine code, but arp packets get handled in software. All these broadcast arps slow this node down noticeably. Is there any particular reason why the bonding code couldn't use unicast arp packets if the arp_ip_target has a valid entry in the sender's arp table? Thanks, Chris -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 2/2] [e1000 VLAN] Disable vlan hw accel when promiscuous mode
David Miller wrote: When you select VLAN, you by definition are asking for non-VLAN traffic to be elided. It is like plugging the ethernet cable into one switch or another. For max functionality it seems like the raw eth device should show everything on the wire in promiscuous mode. If we want to sniff only the traffic for a specific vlan, we can sniff the vlan device. Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
question on sky2 driver panic
Hi, We're using Yukon-XL (0xb3) rev 3 hardware with a vendor-supplied 2.6.14. BAsed on suggestions here, I backported the sky2 driver (v1.10 from 2.6.20.6) to 2.6.14. Unfortunately, when I booted this I got the following: skb_over_panic: text:d00d4e14 len:60 put:60 head:c00264920770 data:c00264920720 tail:c00264920720 end:c002649207a0 dev:NULL kernel BUG in skb_over_panic at /usr/local/src/2.6.14/gd/linux/net/core/skbuff.c:94! Oops: Exception in kernel mode, sig: 5 [#1] SMP NR_CPUS=2 Modules linked in: tipc bond1 bond0 ipmi_devintf ipmi_msghandler NIP: C020D7E8 XER: LR: C020D7E4 CTR: C01C210C REGS: c0025c2aefe0 TRAP: 0700 Not tainted (2.6.14-pne) MSR: 90029032 EE: 1 PR: 0 FP: 0 ME: 1 IR/DR: 11 CR: 28008022 DAR: DSISR: c0025c2af1c0 TASK: cfec2940[2107] 'insmod' THREAD: c0025c2ac000 CPU: 0 GPR00: C020D7E4 C0025C2AF300 C041DA08 009C GPR04: 90009032 0030 C037C428 GPR08: C037EEF0 C043AD68 C043AC88 GPR12: 0010 C0374000 100D47E8 GPR16: 100D55A0 GPR20: 0001 GPR24: B6AA7FFF GPR28: 003C CFEF5B10 C03BB778 003C NIP [c020d7e8] .skb_over_panic+0x50/0x68 LR [c020d7e4] .skb_over_panic+0x4c/0x68 Call Trace: [c0025c2af300] [c020d7e4] .skb_over_panic+0x4c/0x68 (unreliable) [c0025c2af390] [d00d4e20] .named_prepare_buf+0x298/0x2a8 [tipc] [c0025c2af450] [d00d4e90] .named_publish+0x60/0xe4 [tipc] [c0025c2af4e0] [d00d80a8] .nametbl_publish+0x128/0x198 [tipc] [c0025c2af590] [d00de7dc] .tipc_publish+0xe8/0x188 [tipc] [c0025c2af650] [d00d7f4c] .nametbl_publish_rsv+0x30/0x64 [tipc] [c0025c2af6e0] [d00d2600] .cfg_init+0x120/0x150 [tipc] [c0025c2af7a0] [d00e31ac] .process_signal_queue+0xa4/0x100 [tipc] [c0025c2af8/0x1ec [tipc] [c0025c2afcf0] [c00685ec] .sys_init_module+0x28c/0x510 [c0025c2afd90] [c0009b9c] syscall_exit+0x0/0x18 Now granted it looks like this was triggered by tipc, but is there anything that you can think of in the sky2 driver that may have been related? Maybe due to the fragmented buffer handling? The link would have been using an mtu of 9KB. Thanks, Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: question on sky2 driver panic
Stephen Hemminger wrote: Maybe TIPC can't handle fragmented receive buffers. The sky2 driver generates skb's with header and multiple pages if MTU is big enough. For 9K MTU that would be 1K of data + 2 4K pages. The protocols are supposed to be smart enough to handle this, but TIPC is rarely used. We already had to modify tipc to handle fragmented receive buffers when we had memory allocation errors on the e1000 and moved to fragmented skbs in that driver. Our version of the e1000 passes 200 bytes in the initial chunk, and the rest in fragments. tipc currently handles that without any difficulty. I was just checking more to see if there were any known issues in this area that have been fixed in more recent driver versions. Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
sk98lin, jumbo frames, and memory fragmentation
Hi all, We're considering some hardware that uses the sk98lin network hardware, and we'll be using jumbo frames. Looking at the driver, when using a 9KB MTU it seems like it would end up trying to atomically allocate a 16KB buffer. Has anyone heard of this been a problem? It would seem like trying to atomically allocate four physically contiguous pages could become tricky after the system has been running for a while. The reason I ask is that we ran into this with the e1000. Before they added the new jumbo frame code it was trying to atomically allocate 32KB buffers and we would start getting allocation failures after a month or so of uptime. Any information anyone can provide would be appreciated. Thanks, Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: sk98lin, jumbo frames, and memory fragmentation
Stephen Hemminger wrote: Adding fragmentation support to skge driver is on my list of possible extensions. sky2 driver already supports it (yet one more feature that the vendor sk98lin driver doesn't do). Thanks for speaking up. As I mentioned in my email to Jeff it looks like the sky2 driver is what I need (Marvel Yukon 88E8062). However, I'm on 2.6.14 and it doesn't exist there...do you anticipate any issues if I were to backport it? Thanks, Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: sk98lin, jumbo frames, and memory fragmentation
Jeff Garzik wrote: The sk98lin driver is going away, please don't use it. It's unmaintained and full of known bugs. Okay...so it looks like the proper driver for the Marvell Yukon 88E8062 is the sky2 driver, and this one does avoid order0 allocations. Am I on track? Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 0/24] make atomic_read() behave consistently across all architectures
Linus Torvalds wrote: - in other words, the *only* possible meaning for volatile is a purely single-CPU meaning. And if you only have a single CPU involved in the process, the volatile is by definition pointless (because even without a volatile, the compiler is required to make the C code appear consistent as far as a single CPU is concerned). I assume you mean except for IO-related code and 'random' values like jiffies as you mention later on? I assume other values set in interrupt handlers would count as random from a volatility perspective? So anybody who argues for volatile fixing bugs is fundamentally incorrect. It does NO SUCH THING. By arguing that, such people only show that you have no idea what they are talking about. What about reading values modified in interrupt handlers, as in your random case? Or is this a bug where the user of atomic_read() is invalidly expecting a read each time it is called? Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 0/24] make atomic_read() behave consistently across all architectures
Herbert Xu wrote: But I have to say that I still don't know of a single place where one would actually use the volatile variant. Given that many of the existing users do currently have volatile, are you comfortable simply removing that behaviour from them? Are you sure that you will not introduce any issues? Forcing a re-read is only a performance penalty. Removing it can cause behavioural changes. I would be more comfortable making the default match the majority of the current implementations (ie: volatile semantics). Then, if someone cares about performance they can explicitly validate the call path and convert it over to the non-volatile version. Correctness before speed... Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 24/24] document volatile atomic_read() behavior
Segher Boessenkool wrote: Anyway, what's the supposed advantage of *(volatile *) vs. using a real volatile object? That you can access that same object in a non-volatile way? That's my understanding. That way accesses where you don't care about volatility may be optimised. For instance, in cases where there are already other things controlling visibility (as are needed for atomic increment, for example) you don't need to make the access itself volatile. Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: iproute2 showing wrong number of bytes on 64bit architectures.
David Miller wrote: It used unsigned long ages ago, and ifconfig gets the bits exported from /proc/net/dev output whereas we have to used fixed data types in whatever we use over netlink so u32 was choosen. It's rather ironic that the new-and-improved way of doing things is subject to rollover while the old way is not. Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: AF_PACKET how to get the original netdev from a packet received from a bonded master
Chris Leech wrote: Just to give you an idea of our motivation around this, we're looking at layer 2 configuration protocols implemented from user space. I'd like to second the intent of this patch. We've been maintaining a patch against 2.6.10 for a while now that exports the original ingress device to userspace via ancilliary data. We use it in combination with bonding. Was never submitted to mainline because we're on an old kernel. Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
quick help with bonding?
Im doing some experimenting with a new network driver that receives jumbo frames into multiple separate pages that are then joined together in a single sk_buff using skb_fill_page_desc(). It behaved fairly well with standard networking, but its behaving strangely with bonding added to the mix. Could someone either point me to the bonding high level design document (couldn't find one at the sourceforge project page) or else give me a quick overview of the code path followed by an incoming packet when bonding is involved? Thanks, Chris Friesen - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Bonding-devel] quick help with bonding?
Andy Gospodarek wrote: Can you elaborate on what isn't going well with this driver/hardware? I have a ppc64 blade running a customized 2.6.10. At init time, two of our gigE links (eth4 and eth5) are bonded together to form bond0. This link has an MTU of 9000, and uses arp monitoring. We're using an ethernet driver with a modified RX path for jumbo frames[1]. With the stock driver, it seems to work fine. The problem is that eth5 seems to be bouncing up and down every 15 sec or so (see the attached log excerpt). Also, ifconfig shows that only 3 packets totalling 250 bytes have gone out eth5, when I know that the arp monitoring code from the bond layer is sending 10 arps/sec out the link. eth5 Link encap:Ethernet HWaddr 00:03:CC:51:01:3E inet6 addr: fe80::203:ccff:fe51:13e/64 Scope:Link UP BROADCAST RUNNING SLAVE MULTICAST MTU:9000 Metric:1 RX packets:119325 errors:90283 dropped:90283 overruns:90283 frame:0 TX packets:3 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:1000 RX bytes:8978310 (8.5 MiB) TX bytes:250 (250.0 b) Base address:0x3840 Memory:9222-9224 I had initially suspected that it might be due to the u32 jiffies stuff in bonding.h, but changing that doesn't seem to fix the issue. If I boot the system and then log in and manually create the bond link (rather than it happening at init time) then I don't see the problem. If it matters at all, normally the system boots from eth4. I'm going to try booting from eth6 and see if the problem still occurs. Chris [1] I'm not sure if I'm supposed to mention the specific driver, as it hasn't been officially released yet, so I'll keep this high-level. Normally for jumbo frames you need to allocate a large physically contiguous buffer. With the modified driver, rather than receiving into a contiguous buffer the incoming packet is split across multiple pages which are then reassembled into an sk_buff and passed up the link. Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: ARP monitoring set to 100 ms with 2 target(s): 172.24.136.0 172.24.137.0 Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: ARP monitoring set to 100 ms with 2 target(s): 172.25.136.0 172.25.137.0 Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: Warning: failed to get speed/duplex from eth4, speed forced to 100Mbps, duplex forced to Full. Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: enslaving eth4 as an active interface with an up link. Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: Warning: failed to get speed/duplex from eth5, speed forced to 100Mbps, duplex forced to Full. Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: enslaving eth5 as an active interface with an up link. Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: scheduling interface eth5 to be reset in 3 msec. Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: interface eth5 is now down. Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: scheduling interface eth4 to be reset in 3 msec. Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: interface eth4 is now down. Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: now running without any active interface ! Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: cancelled scheduled reset of interface eth5 Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: link status definitely up for interface eth5 Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: cancelled scheduled reset of interface eth4 Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: interface eth4 is now up Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: scheduling interface eth5 to be reset in 3 msec. Mar 29 20:54:08 base0-0-0-5-0-11-1 kernel: bonding: bond0: interface eth5 is now down. Mar 29 20:54:09 base0-0-0-5-0-11-1 kernel: bonding: interface eth4 reset delay set to 600 msec. Mar 29 20:54:59 base0-0-0-5-0-11-1 kernel: bonding: bond0: cancelled scheduled reset of interface eth5 Mar 29 20:54:59 base0-0-0-5-0-11-1 kernel: bonding: bond0: interface eth5 is now up Mar 29 20:54:59 base0-0-0-5-0-11-1 kernel: bonding: bond0: scheduling interface eth5 to be reset in 3 msec. Mar 29 20:54:59 base0-0-0-5-0-11-1 kernel: bonding: bond0: interface eth5 is now down. Mar 29 20:55:15 base0-0-0-5-0-11-1 kernel: bonding: bond0: cancelled scheduled reset of interface eth5 Mar 29 20:55:15 base0-0-0-5-0-11-1 kernel: bonding: bond0: interface eth5 is now up Mar 29 20:55:15 base0-0-0-5-0-11-1 kernel: bonding: bond0: scheduling interface eth5 to be reset in 3 msec. Mar 29 20:55:15 base0-0-0-5-0-11-1 kernel: bonding: bond0: interface eth5 is now down. Mar 29 20:55:30 base0-0-0-5-0-11-1 kernel: bonding: bond0: cancelled scheduled reset of interface eth5 Mar 29 20:55:30 base0-0-0-5-0-11-1 kernel: bonding: bond0: interface eth5 is now up
Re: [Bonding-devel] quick help with bonding?
Jay Vosburgh wrote: 2.6.10 is pretty old, and there have been a number of fixes to the bonding ARP monitor since then, so it may be that it is simply misbehaving (presuming that you're running the 2.6.10 bonding driver). Are you in a position to test against a more recent kernel (and/or bonding driver)? Does the miimon misbehave in a similar fashion? Testing a more recent kernel is problematic. A new bonding driver could be possible, assuming the code hasn't changed too much. I just did another experiment. Normally we boot via eth4 (which then becomes part of the bond with eth5 at init time). If I boot via eth6 instead, it appears as though the problem doesn't show up. Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Bonding-devel] quick help with bonding?
Andy Gospodarek wrote: If you are looking for a decent source for patches you could consider downloading the latest source-rpm from RHEL4/CentOS4. The bonding driver in those releases have been updated to much later code and I can tell you from personal experience they work pretty well. You may need to do some backporting to get the latest arp-monitoring features, but let me know if you need a hand with that, I might have some laying around. ;) I'm just about to load a kernel with a backport of bonding from 2.6.14. I'll try it out and if it doesn't help I'll try the RHEL4 one. Does eth6 use the same hardware/driver as eth4/5? (Sorry if I missed that in the thread, but didn't see if you indicated that it did.) No, eth6 is an AMD-8111. Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Bonding-devel] quick help with bonding?
Chris Friesen wrote: Andy Gospodarek wrote: If you are looking for a decent source for patches you could consider downloading the latest source-rpm from RHEL4/CentOS4. The bonding driver in those releases have been updated to much later code and I can tell you from personal experience they work pretty well. I'm just about to load a kernel with a backport of bonding from 2.6.14. I'll try it out and if it doesn't help I'll try the RHEL4 one. No joy on the 2.6.14 backport, so I guess I'll try the RHEL4 route. Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Bonding-devel] quick help with bonding?
Chris Friesen wrote: No joy on the 2.6.14 backport, so I guess I'll try the RHEL4 route. Bonding driver from 2.6.9-42.0.8.EL doesn't help at all, at least with the module parms I was using before. Switching to miimon doesn't help either. Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [RFC] ARP notify option
Stephen Hemminger wrote: +arp_notify - BOOLEAN + Define mode for notification of address and device changes. + 0 - (default): do nothing + 1 - Generate gratuitous arp replies when device is brought up + or hardware address changes. Did you consider using gratuitous arp requests instead? I remember reading about some hardware that updated its arp cache on gratuitous requests but not gratuitous replies. Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: jumbo frames and memory fragmentation
Herbert Xu wrote: Chris Friesen [EMAIL PROTECTED] wrote: Looking at the page-splitting code, it says 82571 and greater support packet-split We're running the 82546GB device. Looks like it won't help me. Well, time to fork out for a new card then :) I wish. This is an embedded ATCA board. I'm not going to get a respin just because it doesn't deal well with jumbo frames. Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: jumbo frames and memory fragmentation
Herbert Xu wrote: Either upgrade your kernel or backport the page-splitting code in the current tree. That's really the only sane solution for jumbo packets. Looking at the page-splitting code, it says 82571 and greater support packet-split We're running the 82546GB device. Looks like it won't help me. Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: jumbo frames and memory fragmentation
Evgeniy Polyakov wrote: It definitely will. Packet split in hardware means separating data and headers into different pages in different reads, while software page split means that skb has a list of fragments where part of the packet will be DMAed, so jumbo frame will be converted into several pages. Maybe I'm looking at the wrong code then. Can you point me to where this software page split is handled? Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
jumbo frames and memory fragmentation
I'm running a system with multiple e1000 devices, using 9KB jumbo frames. I'm running a modified 2.6.10 with e1000 driver 5.5.4-k2. I'm a bit concerned about the behaviour of this driver with jumbo frames. We ask for 9KB. The driver then bumps that up to a power-of-two, so it calls dev_alloc_skb(16384). That then bumps it up a bit to allow for its own overhead, so it appears that we end up asking for 32KB of physically contiguous memory for every packet coming in. Ouch. Add to that the fact that this version of the driver doesn't do copybreak, and it means that after we're up for a few days it starts complaining about not being able to allocate buffers. Anyone have any suggestions on how to improve this? Upgrading kernels isn't an option. I could port back the copybreak stuff fairly easily. Back in 2.4 some of the drivers used to retry buffer allocations using GFP_KERNEL once interrupts were reenabled. I don't see many of them doing that anymore--would there be any benefit to that? Thanks, Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: Network performance degradation from 2.6.11.12 to 2.6.16.20
Andi Kleen wrote: Incoming packets are only time stamped when someone asks for the timestamps. Doesn't that add scheduling latency to the timestamps? Or is is a flag that gets set to trigger timestamping at packet arrival? Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
a few questions about NAPI
Hi guys, I've got a few questions about NAPI. The help text in the kernel suggests enabling NAPI if incoming packet rate is over 10Kpps. If we're under that rate currently, but could grow above that rate, what would be the impact of enabling NAPI? Why would we not enable it always if available? Are there any gotchas with SMP machines? Are there any other issues of which I need to be aware? Thanks, Chris - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html