sky2 still broken

[l_bratch]

I can confirm it still dies with 2.6.20.

It's very interesting to see people with P5W DH Deluxes saying their machines are freezing - so is mine. It requires a hard reboot when this happens. (no keyboard/mouse respond, and num/caps/scroll locks are all frozen).

Is this related to sky2 do we know?

It seems to be related, according to the bug I linked to and other reports. I've noticed recently that it only seems to happen when I'm away from the computer. When I leave it, it always fails within a couple hours (fortunately, the ext3 and reiserfs journals have prevented any data corruption so far). The NIC will fail, but the kernel keeps running if I'm only away for brief periods. May just be coincidence.

I've noticed exactly the same thing. If I'm on it for a huge amount of time, such as around 20 hours, only coming off for a few minutes at a time, it's fine. But if I leave it over night, or go to work, it crashes around half the time.

Bornio - I shall try that patch tomorrow.

[Bornio]

Bornio, does this work against stable 2.6.20-kernel or do I need newer git-sources ? (I have no access to my rig right now)

thanks

This should work on all recent kernels (2.6.18 and higher I belive).

Please let me know if you have any problems with getting it working.

[l_bratch]

Hi,

Please try the following patch by Stephen Hemminger which was I got from him 4 hours ago.

Code: Select all

--- sky2-dev.orig/drivers/net/sky2.h    2007-02-13 15:08:30.000000000 -0800
+++ sky2-dev/drivers/net/sky2.h 2007-02-13 15:12:52.000000000 -0800
@@ -1589,7 +1589,7 @@

       GMR_FS_ANY_ERR  = GMR_FS_RX_FF_OV | GMR_FS_CRC_ERR |
                         GMR_FS_FRAGMENT | GMR_FS_LONG_ERR |
-                         GMR_FS_MII_ERR | GMR_FS_GOOD_FC | GMR_FS_BAD_FC |
+                         GMR_FS_MII_ERR | GMR_FS_BAD_FC |
                         GMR_FS_UN_SIZE | GMR_FS_JABBER,
 };

this should fix the issues. It also needs broader testing.

Let me know how it goes!

Ok I'm now running with that patch... now we wait.

[TillEulenspiegel]

No change for me, unfortunately. The same problems still occur. I applied the patch on 2.6.20-gentoo.

[l_bratch]

Same here, it had frozen overnight.

Applied to 2.6.20-gentoo too.

[kernelOfTruth]

Bornio, does this work against stable 2.6.20-kernel or do I need newer git-sources ? (I have no access to my rig right now)

thanks

This should work on all recent kernels (2.6.18 and higher I belive).

Please let me know if you have any problems with getting it working.

I will, thanks,

btw.

anyone tried out 2.6.20-git15 ?
it has some interesting additions to sky2, perhaps now it finally works without "hickups"?

[shiver]

I'll try it soon, I need a new kernel anyway since my netfilter config is messed up.

[l_bratch]

I'll try -git15 tonight.

Should I try it with or without the patch Bornio posted?

[serge_g]

Should I try it with or without the patch Bornio posted?

According to http://www.kernel.org/diff/diffview.cgi ... .bz2;z=743
for git-r15 the patch is already applied

[l_bratch]

Well, I'm running with 2.6.20-git15 - I'll report back either in a few days, or when it next dies, whichever comes first

[TillEulenspiegel]

So far, so good.

$ uname -a
Linux localhost 2.6.21-rc1 #1 SMP PREEMPT Wed Feb 21 10:39:11 EST 2007 i686 Intel(R) Core(TM)2 CPU 6600 @ 2.40GHz GenuineIntel GNU/Linux
$ uptime
13:44:35 up 2:52, 2 users, load average: 0.01, 0.03, 0.13

I'll post again tomorrow.

[l_bratch]

Even further, still good:

$ uname -a
Linux bratchc2ddsktop 2.6.20-git15 #1 SMP Tue Feb 20 22:28:53 GMT 2007 x86_64 Intel(R) Core(TM)2 CPU 6700 @ 2.66GHz GenuineIntel GNU/Linux
$ uptime
17:29:02 up 1 day, 18:56, 3 users, load average: 0.04, 0.05, 0.05

I think it's fixed.. rejoice!

[TillEulenspiegel]

The failures still happen, but the driver now resets itself in five minutes. No lockups.

I believe this time (five minutes) can be adjusted by changing line 75 of sky2.c. I'm using:

Code: Select all

#define TX_WATCHDOG             (HZ / 4)

Which should be about 15 seconds. Works as a bandaid for now.
Disclaimer: I am not by any stretch of the imagination a Linux kernel dev. My only practical experience in operating systems is a bit of FreeBSD kernel hacking.

[l_bratch]

I just looked in my dmesg, and you're right:

sky2 0000:04:00.0: v1.13 addr 0xfa9fc000 irq 19 Yukon-EC (0xb6) rev 2
sky2 eth0: addr 00:17:31:f5:a0:01
sky2 0000:03:00.0: v1.13 addr 0xfa8fc000 irq 16 Yukon-EC (0xb6) rev 2
sky2 eth1: addr 00:17:31:f5:9d:63
sky2 eth0: enabling interface
sky2 eth0: ram buffer 48K
sky2 eth0: Link is up at 100 Mbps, full duplex, flow control both
sky2 eth0: tx timeout
sky2 eth0: transmit ring 199 .. 158 report=201 done=201
sky2 eth0: disabling interface
sky2 eth0: enabling interface
sky2 eth0: ram buffer 48K
sky2 eth0: Link is up at 100 Mbps, full duplex, flow control both

It's more usable, but indeed not fixed.

[lonex]

Since our ADSL modem is having issues and looses our Internet connection every few hours, I noted some new things concerning our problem with the sky2 NIC.
With my computer and its sky2 NIC, I'm connected to our router (gigabit). The router is also connected (of course) to the ADSL modem and routes to the Internet. So far so good.
Every time our modem looses its connection while I have some traffic to or from the Internet going on, I have to rmmod and then re-modprobe my sky2 module because I've completely lost my network connection. So I kind of assume that the problem with sky2 is somehow related to tcp-timeouts. Has anyone have seen similar things? Or can re-produce that?

[kernelOfTruth]

confirmed!

I got the same problem with my belkin router, although it's a 100mbit-connection that should make no difference,
I can still login to it's lan / wan ip-number 192.168.2.1 but I got no internet-connection so I have to reboot it

1) the above mentioned is the first variant

2) the second variant is the following:
internet connection is completely lost & I have to rmmod & re-modprobe (as you wrote) the sky2 module, but this only happens on pretty high load, e.g. lots of uploading & then moderate downloading

[gabelhonz]

Hi,

why do you guys do not take the official supported driver?
If you do so you have no problems any more!

Try this:

http://www.marvell.com/drivers/driverDi ... 153&pId=36

it's always up2date and works from kernel 2.4.20 to 2.6.latest.
Compile as module for all options to work!

regards

gabelhonz

[ttuegel]

I don't think the Marvell official driver supports their Yukon2 chipset, which is what we're all using. There is a driver for that chip from SysKonnect. However, work is finally progressing on the in-tree driver and should continue to progress, as long as people are testing the driver.

[gabelhonz]

My friend,

in this Driver Release ALL Yukon Devices are supported.
Yukon, Yukon2, YukonEC .......
Support for all NICS in one Driver !!

It doesnt matter if you download the sources from www.marvell.com or www.skd.de!!! It's the same.
Check the source code and you will see.

In this package sk98lin loads all SysKonnect and Marvell NICs! skge and sky2 are modified Driver from Mr. Hemminger.
YOU SHOULD USE THE OFFICAL DRIVER PACKAGE!!!

EDIT: Only with the official Driver you can use eg. RLMT on Dual Port Yukon2
in skge and sky2 you have only basic functionality of your device.

regards

[ttuegel]

I see, I didn't realize that the SK driver was available on Marvell's site now, too.

This is, of course, good for people who need a fully working driver _now_, either because they run a server, or don't have another NIC, but hopefully folks here will continue to test the in-kernel drivers.

4. NO SUPPORT. Marvell may make changes to the Deliverables at any time without notice. Nothing in this Agreement obligates Marvell to support or provide updates or error corrections to the Deliverables.

These sorts of license agreements come as no surprise, but I've had drivers break due to kernel upgrades before... Yeah, Marvell will keep the drivers up to date, I'm sure, bumping the driver version to keep up with the kernel until something actually breaks.

[l_bratch]

I'm getting freeze again with the in kernel driver, this is the same version that seemed to have fixed the freezing by just disabling the card for a time, yet has gone back to freezing.

I'm going to give the Marvell one a try. (Because I have no other network card, but I'll try the in kernel one again every time I see a change to it in the changelog )

Edit:
I'm not using the Marvell driver, I'll post back with how I get on...

Output from dmesg if anyone is interested:

sk98lin: Network Device Driver v10.0.4.3
(C)Copyright 1999-2007 Marvell(R).
PCI: Setting latency timer of device 0000:04:00.0 to 64
eth0: Marvell Yukon 88E8053 Gigabit Ethernet Controller
PrefPort:A RlmtMode:Check Link State
ACPI: PCI Interrupt 0000:03:00.0[A] -> GSI 16 (level, low) -> IRQ 16
PCI: Setting latency timer of device 0000:03:00.0 to 64
eth1: Marvell Yukon 88E8053 Gigabit Ethernet Controller
PrefPort:A RlmtMode:Check Link State
ADDRCONF(NETDEV_UP): eth0: link is not ready
eth0: network connection up using port A
speed: 100
autonegotiation: yes
duplex mode: full
flowctrl: symmetric
irq moderation: disabled
tcp offload: enabled
scatter-gather: enabled
tx-checksum: enabled
rx-checksum: enabled
rx-polling: enabled
ADDRCONF(NETDEV_CHANGE): eth0: link becomes ready
eth0: no IPv6 routers present
eth0: network connection down
ADDRCONF(NETDEV_UP): eth0: link is not ready
eth0: network connection up using port A
speed: 100
autonegotiation: yes
duplex mode: full
flowctrl: symmetric
irq moderation: disabled
tcp offload: enabled
scatter-gather: enabled
tx-checksum: enabled
rx-checksum: enabled
rx-polling: enabled
ADDRCONF(NETDEV_CHANGE): eth0: link becomes ready
eth0: no IPv6 routers present

[lenk]

Hi, I still get the occasional hang when downloading torrents. Its not too bad. Running 2.6.20-viper4.

There is a more pressing problem. Do anyone here run vmware workstation? I've got Windows XP running as a guest in vmware workstation. The problem is that I can copy files using winscp from the guest (winxp) to my gentoo box (SSHD) but I CAN'T copy files from the host (gentoo) to guest. The guest is bridged to the host nic. I've tried this exact same setup on my lappy and it works perfectly. I have to assume its the sky2 drivers fault. I'll get a cheapo nic tomorrow to confirm.

--update: I've tried the driver from marvell, same issue.

[lenk]

Just bought a new 10/100 card and networking from host to guest in vmware bridged mode works. Its the sky2 driver that is broken.

[PaulBredbury]

Here's an ebuild for Marvell's driver. It works with kernels 2.6.20 upwards.

/usr/local/portage/net-misc/sk98lin/sk98lin-10.21.1.3.ebuild

Code: Select all

# Copyright 1999-2007 Gentoo Foundation
# Distributed under the terms of the GNU General Public License v2
# $Header: $

# http://forums.gentoo.org/viewtopic.php?p=3946286#3946286

inherit eutils linux-info

DESCRIPTION="sk98lin kernel driver for Marvell network cards"
HOMEPAGE="http://www.marvell.com/drivers/driverDisplay.do?dId=153&pId=36"
SRC_URI="http://www.marvell.com/drivers/upload/install_v${PV}.tar.bz2"

LICENSE="as-is"
SLOT="0"
KEYWORDS="x86"
IUSE=""

DEPEND="virtual/linux-sources"

S=${WORKDIR}/DriverInstall

src_compile() {
	# Create patch
	./install.sh -p /usr/src/linux ${PN}.patch
}

src_install() {
	dodoc README
}

pkg_postinst() {
	# Guard against applying patch twice.
	if [[ -n $(grep "Driver for Marvell Yukon/2 chipset" /usr/src/linux/drivers/net/sk98lin/sky2.c) ]] ; then
		die "sk98lin is already installed for this kernel."
	fi

	# Apply patch. Runs outside of the sandbox.
	patch -d /usr/src/linux -p1 < "${S}"/${PN}.patch \
		|| die "patch failed"

	if kernel_is ge 2 6 23 ; then
		# http://forums.gentoo.org/viewtopic-t-487018-start-275.html
		elog "Patching eth_copy_and_sum."
		sed -i \
			-e "s:eth_copy_and_sum(pNewMsg, pMsg->data,:skb_copy_to_linear_data(pNewMsg, pMsg->data,:" \
			-e "s:FrameLength, 0);:FrameLength);:" \
			/usr/src/linux/drivers/net/sk98lin/skge.c || die "sed skge.c"
	fi

	echo
	elog "The kernel has been patched, to *replace* module sk98lin."
	elog "CONFIG_SK98LIN_NAPI is a new option, provided by this patch."
	elog
	elog "Select sk98lin in 'make menuconfig' and run:"
	elog "   cd /usr/src/linux && make clean bzImage modules modules_install"
	echo
}

Edit: Added eth_copy_and_sum patch, to compile with kernel 2.6.23.1

[Interimo]

I have a question about all of this. I have the revision 22 Marvell NIC (Don't have the exact specs, I'm on Windows as it's the only OS I have that can access the internet) on the Gigabyte GA-965P-DS3 mobo. During the installer, I would frequently need to exit out of the chroot environment and run net-setup again. The install is done, and now there is no net-setup for me to run. I wouldn't think this was a problem, because dhcpcd sets up my network (or so I thought) and everything is supposedly in working order. ifconfig brings back good results, but I can't emerge anything at all. I have sky2 loaded, and removing + re-modprobing doesn't help either. I was thinking about using one of the drivers posted earlier, but since I can't download anything on Gentoo, is it possible to get it off my Windows partition and onto Gentoo's? I already have ntfs-3g installed, so read/write is no problem (I think..), but I don't know how I would go about getting it from one to the other and then proceeding with the install.

Any help would be greatly appreciated.