Laying To Rest That Odd Linux Kernel Regression

[Adarion]

@darkfires

No, sometimes it is just that Linux finally exposes hardware bugs to the public. Like these UEFI bricking of boxes. Linux just does things as they were meant to be, according to protocol. But suddenly the HW gets a hiccup. And then devs bisect the problem and find out that the HW does not do what it is supposed to in the specs.

Anybody remember LG? flush cache atapi command was not flushing the r/w cache of the drive - as it was standardized. No, LG's drives flushed their firmware and were bricked. Ouch.

[bms20]

Hardware troubles

Also, don't underestimate the ability for board vendors to cock up the UEFI. (Although cock up may be too harsh: release prematurely is probably better).

I spent two months analyzing a problem with one baytrail platform - the vendor then released an update which has pretty much resolved it now.

The best symptom of the problem is that I could find the kernel ignoring spinlocks - that's how I concluded it wasn't something I could fix.

The problem manifested itself is many many other horrible ways: Disk corruption, network packets not being delivered, random null pointer crashes in the scheduler, network stack and (rarely) graphics drivers.

-bms

[Jabberwocky]

Can't stop ... writing...

TL;DR: 1) Companies sometimes target other OSs (Nvidia Optimus ), Linux is not to blame for their lack of support. 2) Stay of the bandwagon. 3) Buy good hardware. 4) Memtest is not a system debugger

I've seen lots of weird kernel crashes lately since 3.12, very random... lockups and freezes on various different hardware. Linux kernel quality (more specifically, reliability) is really going down hill if you ask me. Won't even go into how Linux desktop has been going down hill since 2005...

I acknowledge that kernel reliability is not as good as it can be. We have seen various regressions over the years, but most of those problems were related to newer kernels.

Regarding the desktop I don't agree with you at all. I have been using Linux since 2003 thinking back of installing early AMD64 Gentoo using a single PC with 56k modem and a printed how-to manual. Compiling Mandrake 9.2 kernel modules sucked into RPM hell to get 2d/3d acceleration. Installing Ubuntu dapper drake on most of my friends' computers (windows viruses help me persuade them) and helping them "upgrade" to edgy that made those with Geforce 6000 series cards to go back to windows permanently (despite loving Amarok so much ). I really don't know how you can say it has gone down hill. Look at Intel and AMD's open source drivers and support there are so many things that have improved so much you can't even compare. The only negative thing for me is DE fragmentation and user friendliness that has not grown as faster as other OSs, but it still improved a lot. It just did not grow as fast as others. I say this again, we have had some growing problems... like Pulse Audio, but again only for those on the bandwagon.

Two machines at work both 3.12.35 were randomly crashing, the crash dumps appeared to be network related.. Swapped the hardware and also upgraded the kernel to 3.14, same issues though the crash traces were a little different. Memtest was clean on both machines.

I ended up "working around" the issue (by mistake), snmpd was logging every tcpwrapper connect to it via syslog (this by itself isn't really an issue but pointless filling of logs), and disabling that seemed to stop the kernel crashes... makes no sense to me. I did not try downgrading the kernel tho, maybe I should have..

I still have x2 biostar A785GE and x1 GF8100 M2G+ boards stashed somewhere because nobody wanted them. Those network cards only worked to an extent in windows vista, not windows xp, windows 7 nor Linux. Just because Realtek screwed up big time does not mean Linux is to blame. There's a reason why VMware ESX dropped ALL Realtek NICs.

Disabling the logs proves nothing. If you don't know what the cause of the problem is you can't put blame on something.

Another system , brand new Z97 setup randomly locks up after a few days, not even a freaking message or anything.... memtest clean on this as well..... sigh.

In other news, my Windows 10 machine has a 4week uptime; works great...

I miss the days when nothing crashed Linux and it was windows puking it's guts out... funny how times change.

I think most of us can agree that Windows has improved a lot over the past few years and it is funny how times change, hell look at the dev version of I.E. 11 it has better ES6 support than Chrome or Firefox! Has Windows improved enough to get some of us Linux users to move (back) to it... I don't know... If you don't have any problems on Windows 10 you're the luckiest person on the planet, but the day that you have a problem and there's no way of finding out what is wrong with your system and all you can do is helplessly do memtests until your face turns blue... I think then you might start to miss Linux ;-) Yes sometimes we have to get our hands dirty, but we have an open source OS you can look at every piece of code and you always have a good community helping us when we get stuck (if you don't buy bad hardware). Until Windows has that I won't be moving back it does not matter how bad companies support Linux or how unstable newer releases are (supposedly).

It's funny how leading manufacturers really **** up new hardware. From my personal experience Nvidia, Dell and Gigabyte.

[Xaero_Vincent]

Yeah I can defintely say Linux crashes more so than Windows 7 or 8.1. That's not to say Windows is really better--you have viruses and malware to deal with (I know I deal with this almost every day at work being a PC tech and it still happens even if the customer has anti-virus software--granted this mostly has to due with Windows 90+% marketshare) and also my Windows system feels very sluggish when when booting. Linux (at least Archlinux) runs and boots faster for me.

In any case, I think my crashing has to do with the proprietary AMD graphics drivers. I can get my system to crash by simply switching between TTYs a few times--the keyboard and mouse freezes and then after a few seconds my monitor goes into standby and I'm forced press the reset button. Just the other day I re-enabled "Ctrl-Alt-Backspace" (Kill X shortcut) and also created a shortcut key "Ctrl-Alt-\" to kill Cinnamon Session but not programs running (because Cinnamon sometimes freezes up). So I'm hoping with this I can avoid using TTYs to kill misbehaving apps that freeze the desktop and prevent me from using my computer.

As far as the kernel goes...it's pretty solid for me but Linus needs to put his foot down and say "hey... lets do a bug fix point release--no new features just fix bugs." The truth is the Linux kernel development is occuring at breakneck pace and with it new bugs are introduced just as quickly. I can tell Archlinux devs are more cautious about pushing kernel upgrades to "Core" than I remember in the past. The lastest is still 3.17.6, yet 3.18 has been released a month ago and still in testing.

I think some of us put up with these problems in the name of freedom. The biggest point and advantage of using Linux and FOSS in general is the 4 freedoms Richard Stallman talks about. The others are more control over your system, far less malware or almost none and faster system performance. But Windows is clearly no longer the unstable BSOD crap OS it once was and I still use Windows 8.1 on one of my laptops, mainly to use Steam In-Home Streaming and play some DirectX10 games on my Linux machine that don't work in Wine.

[gamerk2]

Xaero, I'll actually go one step farther: I'd argue that as of Vista SP1/7, a clean Windows install is more stable then a clean Linux install. Say what you will about Microsofts design choices, the Windows kernel, on the whole, is very stable now.

Seriously, I had to try and find a way to break the OS for non-hardware failures. Only managed to do so once, and it was a real oddball case that got fixed back in Vista SP1.

[curaga]

Xaero, I'll actually go one step farther: I'd argue that as of Vista SP1/7, a clean Windows install is more stable then a clean Linux install. Say what you will about Microsofts design choices, the Windows kernel, on the whole, is very stable now.

Seriously, I had to try and find a way to break the OS for non-hardware failures. Only managed to do so once, and it was a real oddball case that got fixed back in Vista SP1.

If a clean install means no additional software, it isn't quite fair unless on the linux distro you only run the Notepad equivalent

[NotMine999]

I think most of us can agree that Windows has improved a lot over the past few years and it is funny how times change, hell look at the dev version of I.E. 11 it has better ES6 support than Chrome or Firefox! Has Windows improved enough to get some of us Linux users to move (back) to it... I don't know... If you don't have any problems on Windows 10 you're the luckiest person on the planet, but the day that you have a problem and there's no way of finding out what is wrong with your system and all you can do is helplessly do memtests until your face turns blue... I think then you might start to miss Linux ;-) Yes sometimes we have to get our hands dirty, but we have an open source OS you can look at every piece of code and you always have a good community helping us when we get stuck (if you don't buy bad hardware). Until Windows has that I won't be moving back it does not matter how bad companies support Linux or how unstable newer releases are (supposedly).

It's funny how leading manufacturers really **** up new hardware. From my personal experience Nvidia, Dell and Gigabyte.

Windows may have gotten more stable as people that have migrated up to Windows 7 have numerous Microsoft Windows "Customer Experience Improvement Program" (CEIP) tasks quietly hidden within various directories and the Windows Task Scheduler. If you forget to "opt out" of CEIP, then multiple tasks become "active" in Task Scheduler and they periodically "phone home" to Microsoft with all sorts of information regarding various aspects of the machine and installed Windows OS.

Now placing the "tin foil hat" aside, I think the Windows user community owes some thanks to those people that never "opted out" of CEIP and contributed their data to Microsoft. The same thing can be said about people that allow "Windows Error Reporting" (WER) to report various errors to Microsoft and with a minimum of user effort, compared to filing a bug report in the Linux community; even Bruce Byfield over at "Linux Pro" magazine and written online about the hassles of Linux bug reporting. So I guess in some ways there are some things that the Linux community could (re)learn(?) from the Windows community, but someone has to want to develop & test such tools, then provide them to the community, not to mention provide the necessary funding to launch and maintain such an effort.

As for my own HW issues, I don't like Dell either; HP is my preference over Dell. Nvidia & AMD/ATI based video cards have never caused me issues, but I don't "push them hard" either. And don't get me started on Gigabyte boards; I know some people that swaer by them, but I have only sworn at them. I will stick with Asus for Windows workstations and Supermicro for my server boards & "high end needs", but I am undecided on a preferred vendor board for Linux based workstations. I have tried in the past: Foxconn (hit & miss, mostly miss), Intel (sometimes that is Foxconn assembled or sourced, so hit & miss...about 50-50), ECS (pa..leeze!!), Tyan (sometimes quirky), and a few others.