Bug 41322 - [Sandy Bridge] e1000 IRQ being disabled resulting in 2MB/s network speeds.
Summary: [Sandy Bridge] e1000 IRQ being disabled resulting in 2MB/s network speeds.
Alias: None
Product: Drivers
Classification: Unclassified
Component: Platform (show other bugs)
Hardware: All Linux
: P1 high
Assignee: drivers_platform@kernel-bugs.osdl.org
Depends on:
Reported: 2011-08-18 02:43 UTC by Simon Lea
Modified: 2012-08-30 10:37 UTC (History)
8 users (show)

See Also:
Kernel Version:
Tree: Fedora
Regression: No


Description Simon Lea 2011-08-18 02:43:37 UTC
Description of problem: I have just upgraded from an Intel C2D 775 processor
and MB to a i3(2100) 1155 processor and MB and have been seeing terrible
network speeds and sometimes terrible disk to disk transfer speeds.  This is
affecting both Fedora 14 and Fedora 15.  I use an Intel GT 1000/Pro network
card rather than the motherboard NIC.

On further investigation of the system log (/var/log/messages) I found the

irq 16: nobody cared (try booting with the "irqpoll" option)
Pid: 0, comm: swapper Not tainted #1
Call Trace:
<IRQ> [<ffffffff810a74f3>] __report_bad_irq.clone.1+0x3d/0x8b
[<ffffffff810a765b>] note_interrupt+0x11a/0x17f
[<ffffffff810a813b>] handle_fasteoi_irq+0xa8/0xce
[<ffffffff8100c2ea>] handle_irq+0x88/0x90
[<ffffffff81470b44>] do_IRQ+0x5c/0xb4
[<ffffffff8146b093>] ret_from_intr+0x0/0x11
<EOI> [<ffffffff8102b7dd>] ? native_safe_halt+0xb/0xd
[<ffffffff81290b75>] acpi_safe_halt+0x2a/0x43
[<ffffffff81290bae>] acpi_idle_do_entry+0x20/0x30
[<ffffffff81290c27>] acpi_idle_enter_c1+0x69/0xb6
[<ffffffff8146e01e>] ? notifier_call_chain+0x14/0x63
[<ffffffff81395d56>] ? menu_select+0x177/0x28c
[<ffffffff81394d6d>] cpuidle_idle_call+0x8b/0xe9
[<ffffffff81008325>] cpu_idle+0xaa/0xcc
[<ffffffff81452876>] rest_init+0x8a/0x8c
[<ffffffff81ba1c49>] start_kernel+0x40b/0x416
[<ffffffff81ba12c6>] x86_64_start_reservations+0xb1/0xb5
[<ffffffff81ba13c2>] x86_64_start_kernel+0xf8/0x107
[<ffffffff813148a8>] (ata_bmdma_interrupt+0x0/0x1a) <-- Via PATA controller ??
[<ffffffffa00d8622>] (mpt_interrupt+0x0/0x890 [mptbase]) <--Sata PCIe
[<ffffffffa00d8622>] (mpt_interrupt+0x0/0x890 [mptbase]) <--Sata PCIe
[<ffffffffa0132850>] (e1000_intr+0x0/0xe9 [e1000]) <--- Network Card
Disabling IRQ #16 

After turning off the audio, USB3 and network controller on the MB I still got
the same error (only the e1000 affected this time and a different IRQ number). 
The error can occur at any time, not just on boot, but it usually happens
fairly quickly.  The ata_bmdma_interrupt and e1000_intr were always affected.  I cannot turn off the PATA controller without disabling the onboard SATA ports which are needed.

Version-Release number of selected component (if applicable):
Fedora 14 - ( 
Latest Fedora 15 install CD (as of last week).

How reproducible:

Steps to Reproduce:
1. Turn server on in above config.
2. Perform a large file transfer (5-10GB)
3. Watch for network transfer speed to change.

Actual results:
Network transfer speeds not able to go higher than 3MB/s
Disk to disk transfer speeds also sometimes affected depending on which devices
were also using the same IRQ which was disabled.

Expected results:
Normal boot, network speeds of approx 80-90MB/s (obtained when using the
previous Intel C2D MB and CPU).  Running Windows Server 2008r2 (SBS2011 / WHS 2011) exhibits no issues so the hardware would appear to be sound.

Additional info:

Hardware list;
Intel i3 2100 CPU (LGA1155)
Asus P8H67-V (v3) MB
4Gb DDR3 Ram 
2xLSI 1068e PCIe (8 sata drive controllers)
1xIntel GT 1000/Pro (PCI) network controller.
Various hard drives
Norco 4220 case.

Video is via the i3 and motherboard.

The box is configured as a NAS and so no sound chipset is required.

Originally reported in RedHat Bugzilla -> https://bugzilla.redhat.com/show_bug.cgi?id=713351

A quick search here has highlighted an number of other people with the same or very similar issues.

I do not have Linux installed on this machine as it is a functioning server and needs to work but am interested in getting Linux back on it to use as a virtulization platform (have upgraded to an i5 and 12GB ram).  This is not possible if this bug has not been rectified.
Comment 1 Andrew Morton 2011-08-19 22:22:47 UTC
Thomas told me:

> Weird, that this shares interrupts It's PCIe, so it should use MSI and
> not a shared interrupt on the IOApic. We'd need a config and a full
> dmesg to see what's possibly going on.

Note You need to log in before you can comment on or make changes to this bug.