Bug 9648
| Summary: | linux-image-2.6.23-1-amd64: kernel Oops, (segfault) when downloading from net (sky2 driver) | ||
|---|---|---|---|
| Product: | Drivers | Reporter: | Ondrej Certik (ondrej) |
| Component: | Network | Assignee: | Stephen Hemminger (stephen) |
| Status: | RESOLVED INSUFFICIENT_DATA | ||
| Severity: | high | CC: | alan, bunk, jgarzik |
| Priority: | P1 | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Kernel Version: | 2.6.24-rc6 | Subsystem: | |
| Regression: | No | Bisected commit-id: | |
|
Description
Ondrej Certik
2007-12-27 14:38:11 UTC
I should stress, that I tried 2.6.24-rc6 and it still fails. More info in the Debian bug. How much memory do you have? My guess is that problem is memory mapping above 4G. Some hardware doesn't work, it may not even be a device driver issue but more of a DMA mapping problem. Good guess, I have 4GB:
$ free -om
total used free shared buffers cached
Mem: 3962 1849 2113 0 39 1660
Swap: 3820 0 3820
I think I have exactly 4GB, but I am no expert in this.
Actually very nice guess. The i386 worked because it wasn't compiled with big memory support (-bigmem Debian package). And actually, recently I installed the -bigmem i386 Debian kernel package and I experienced sudden hangs - but there was nothing in logs, so I thought it's because of running i386 on 64bit hardware, so I installed Debian amd64 on the box with true amd64 kernel. And the issue just became more apparent. With 4G the BIOS has to "make a hole" and move some of the memory above 4G. In order to reach that memory, some translation is needed, or the system has to avoid that memory for DMA access. What motherboard is this? I see. The motherboard is: Gigabyte GA-965P-DS3 - Intel P965 I have that motherboard, and was using 4G of memory without problem. Check your BIOS version, they are up to F11. http://america.giga-byte.com/FileList/BIOS/motherboard_bios_ga-965p-ds3_f11.exe Is there a way how to get this information remotely over ssh? But currently it doesn't matter, I cannot log in anymore $ ssh -vvv august [...] debug1: SSH2_MSG_KEX_DH_GEX_REQUEST(1024<1024<8192) sent debug1: expecting SSH2_MSG_KEX_DH_GEX_GROUP Connection closed by 147.32.50.195 the kernel probably shot itself in the head again. I'll be sitting at this computer again on Wednesday, so I'll tell you then. While I'll be there, is there something else I can check? I managed to login: $ sudo dmidecode # dmidecode 2.9 SMBIOS 2.4 present. 39 structures occupying 1198 bytes. Table at 0x000F0100. Handle 0x0000, DMI type 0, 24 bytes BIOS Information Vendor: Award Software International, Inc. Version: F4 Release Date: 01/08/2007 [...] So it seems the version is F4, while the newest one seems F11. But I this computer is new, also the Release Date is recent, so the "dmidecode" is maybe lying. So the problem was in the sky2 driver or hardware. I installed the latest released 2.6.24 kernel, which is in Debian, I tried both 64 bit and 32 bit with low mem, and I can reproduce the problem 100% now (no matter if 64bit, 32bit, high mem, low mem) - I just need to download 50MB+ from the net and it hangs for sure. I got fedup with this, so today I took an old RealTek RTL8139 network card, put that in, switched the ethernet cable and it works like a charm. I stressed the computer a lot today, I downloaded 700MB+ from the net, upgraded hundrends of packages in Debian, built 200MB of C++ sources in parallel (on all four processors) several times and I haven't experienced a single hang. And I run on the 64 bit kernel + Debian. So my problem is fixed now, but the real problem is imho either in the sky2 driver, or hardware. I am not an expert in either, so unless someone will be directing me what to do, I cannot help much. As to the bios - I didn't see the version of it at boot or in the bios setup, so all information I can get is given in my previous posts. Ondrej Closing old stale bugs,please re-open if still present in 2.6.29+ |