Latest working kernel version: Any known.
Earliest failing kernel version: 2.6.27-11-generic
I've an Imedia 8783, with a athlon 64 x2 4400+, a Canberra motherboard from them :
Name : Canberra
Type : Micro ATX motherboard
Made by : ECS
* Socket AM2+/AM2 (AMD 940-pin)
* Support socket AM2+/AM2 for AMD Phenom, Athlon 64 X2, Athlon 64, and Sempron processors
* Support HyperTransport 3.0 with backward compatibility to HyperTransport 1.0
* NVIDIA GeForce 8200 mGPU (MCP78)
* 4x 240-pin DDR2 SDRAM DIMM slots
* Support Dual channel configuration for better performance
* Support for single-sided or double-sided DIMMs (DDR2 533/DDR2 677/DDR2 800/DDR2 1066)
* Support for up to 8 GB system memory
* Around 700 MB upper memory is shared with the system when 4 GB RAM is installed
* Maximum video memory in BIOS is 256 MB shared
* Total maximum video memory 1407 MB using HyperMemory
* Realtek ALC 888 audio codec
* High Definition Audio subsystem
* 7.1 channel support
* Realtek RTL8101E
* 10/100 Mbit
* 1x PCI Express 16X
* 1x PCI Express 1X
* 2x PCI
* 4x SATA II
Problem Description: Can't boot using a live cd, can boot using a bootable usb key, but no sata peripheral recognised, like the hard drive, the dvd drive (unlike usb ones which are)...
Steps to reproduce: just try to boot...
Already reported at Launchpad : https://bugs.edge.launchpad.net/ubuntu/+source/linux-meta/+bug/304505, please read here for more details and especially output files like dmesg, kernel.log...
Does "pci=nomsi" make any difference?
No, it doesn't make any difference... :/
Sorry, I spoke a little bit too fast... It works !!!
But should I keep this option for ever or what ? And where was the problem ?
Thanks for your help !!
Created attachment 20066 [details]
output of lspci -vv, if you need it
Can you please do "lspci -nnvv"?
Created attachment 20199 [details]
Here is the result of lspci --nnvv, launched as root
Here is the result of lspci -nnvv, launched as root. Sorry for the late reply, I just went now back on the computer. the result of the same command as standard user follow.
Created attachment 20200 [details]
lspci -nnvv as standard user
Peer Chen, ahci is failing on mcp78s due to broken MSI. Can you please tell me whether MSI should be disabled on the whole system or disabling it on ahci is enough? ie. Is MSI broken in the bridges or is it the ahci controller itself?
I've got a user on Ubuntu, using both the Jaunty alpha 2.6.28- and our packaged vanilla mainline 2.6.29- kernel, that is suffering 'lost disk on resume' syndrome with the Nvidia MCP78S [10de:0ad0] unless the "pci=nomsi" work-around is specified.
We have some comprehensive logs and the user is keen to help in resolving the issue properly.
Hey, tj, I'm a tj too. Nice to meet you. :-)
I'll ping pchen via email one more time and then just blacklist the whole system. Hmm... if anyone is willing to, this can be verified by putting in a MSI capable device and see whether it works without pci=nomsi. e1000e is a good test card for this purpose.
we have a patch for MCP78 MSI issue and kernel 2.6.28 had include it.
The patch code is in drivers/pci/quirks.c and looks like kernel resume code doesn't invoke the patch routine in quirks.c.
Thanks for the pointer Peer Chen. Nicolas, Tj, can you guys please try the patch? Adding call to it during resume path can be easily done adding another fixup declaration, so no worries there.
Is the patch included by default ? I tried to delete the "pci-nomsi" option at booting one time using 2.6.28-11.36 kernel in Jaunty (devel version), and got the same problem that I had, and fell down to a busybox.
I was meaning the "pci=nomsi" option of course...
Sorry for the mistake.
The patch referred to by Peer Chen looks to have got lost in the noise. That may be because it appeared as a 'reply' to another email.
Is there an upstream commit we can cherry-pick? If not, I'll propose the patch for inclusion as a temporary fix until upstream receives it.
The upstream commit is 9dc625e72309e1c919ea3e7f51d0ffca96123787. Is someone interested in testing suspend/resume and whether RESUME is necessary there too?
OK, found it now. That commit is in the Ubuntu kernels since v2.6.25-rc3 (Intrepid 8.10 and Jaunty 9.04).
So the concern is that it isn't being invoked on resume?
Any kernel patch to add the call to quirks code in resume path?
Created attachment 21284 [details]
Sorry about the delay. Patch attached.
Thanks, the patch works, could you help to submit it?
Patch submitted. Thanks.
Thanks for that Tejun. I'll try to get it pulled into the existing Ubuntu releases as well as the forthcoming Karmic release.