Re: [Cooker] mem=nopentium and AMD
baal == Baal [EMAIL PROTECTED] writes: Hi baal The installer could check for the kind of CPU and adjust in baal consequence... As it is not nown a good pattern to recognize the machines affected (normally they are AMD + VIA + NVidea, but not all). That means that only good solution for the insntaller is trying to stress AGP too hard, and if the machine crashes, put mem=nopentium. Ooops, but if we have crashed, we can't change anything :( Later, Juan. PD. Pixel, solution to this problem is easy: Installer in OpenGL now :) -- In theory, practice and theory are the same, but in practice they are different -- Larry McVoy
Re: [Cooker] mem=nopentium and AMD
On Tue, Feb 26, 2002 at 02:09:03PM -0500, David Walluck wrote: be held responsible for *commercial* drivers though (What's with the kernel complaining about the licenses though? Is it illegal to have non-GPL modules? I think this is a patch that was added to the mdk kernel.). Anyway, I like the speed increase of the NVIDIA drivers, but the stability should come first. David, This was discussed ages ago. IIRC it's not a Mandrake patch but something that was added upstream to the kernel. I guess if you don't add some flag to your module it whines. The intent here was to keep commercial module complaignts and bugs off the linux kernel mailing lists. -- Ben Reser [EMAIL PROTECTED] http://ben.reser.org What difference does it make to the dead, the orphans, and the homeless, whether the mad destruction is wrought under the name of totalitarianism or the holy name of liberty and democracy? - Ghandi
Re: [Cooker] mem=nopentium and AMD
On Sat, 2002-03-02 at 01:39, Ben Reser wrote: On Tue, Feb 26, 2002 at 02:09:03PM -0500, David Walluck wrote: be held responsible for *commercial* drivers though (What's with the kernel complaining about the licenses though? Is it illegal to have non-GPL modules? I think this is a patch that was added to the mdk kernel) Anyway, I like the speed increase of the NVIDIA drivers, but the stability should come first David, This was discussed ages ago IIRC it's not a Mandrake patch but something that was added upstream to the kernel I guess if you don't add some flag to your module it whines The intent here was to keep commercial module complaignts and bugs off the linux kernel mailing lists Alan Cox email explaining this: http://lwnnet/2001/0906/a/ac-licensephp3 =--= kk1
Re: [Cooker] mem=nopentium and AMD
Le Mardi 26 Février 2002 15:04, Robert Fox a écrit : On Tue, 2002-02-26 at 14:43, Pixel wrote: [EMAIL PROTECTED] (Robert Fox) writes: [...] My machine has a nasty tendancy to hard crash when I play BZFLAG for extended periods without the mem=nopentium option. I also read somewhere that this options if highly recommended (for stability) for AMD based mahines. Why doesn't the DrakX install catch this and do this automatically? because I really think it should be fixed in the kernel? if mem=nopentium is *needed*, why not have it by default! Interesting argument - if we wait long enough, most things will be fixed in the kernel (eventually). The problem simply is the trivial differences between AMD and Intel based machines. Even th GCC compiler has AMD specific compiling optimizations. Some have argued that we should also have an AMD optimized kernel . . . . Bottom line is - my machine crashes HARD without this option - and I would venture to guess there are several others out there who also have AMD based machines - and some of these people may not be technical enough to find and/or install this option in lilo.conf - too hard for a newbie. Agreed it should be fixed in the kernel - but how long can we afford to wait - and what are the consequences of NOT adding the option for the next (imminent) Mandrake release?? Cheers, R.Fox It seems that i am encountering this problem from time to time. (when i really use the gl display). how did you trace the problem, to make sure it si related to the same problem ? Stef
Re: [Cooker] mem=nopentium and AMD
Le Mardi 26 Février 2002 21:13, Baal a écrit : On Tue, 2002-02-26 at 12:47, David Walluck wrote: Robert Fox wrote: Agreed it should be fixed in the kernel - but how long can we afford to wait - and what are the consequences of NOT adding the option for the next (imminent) Mandrake release?? Actually, there seems to be a bug in drakboot, where there is mem=nonpentium (By default? why?), but when you go to click OK, it says that mem must be a number. Obviously, it doesn't have to be a number. nopentium was probably written there because you performed an upgrade from a system that already had the option appended to LILO... I think it would be very important to add this since it affects the majority of AMD cpus used with the AGP port... The change that mem=nopentium does is simply telling the kernel to use 4k pages instead of 4M pages and almost no programs can use 4M pages anyway so the performance loss is absolutely imperceptible. Many person reported big stability problems when 8.1 came out and this option can prevent that... For those interested in knowing if their CPU is affected, you can make cat /proc/cpuinfo and if the tree following lines are not identical then your CPU probably has the bug: cpu family : 6 model : 2 stepping: 2 Regards, Baal Is it realted only to this CPU's (from where do you take that values), i'm seeing it and here is what provies cat /proc/cpuifo : processor : 0 vendor_id : AuthenticAMD cpu family : 6 model : 6 model name : AMD Athlon(tm) XP 1600+ stepping: 2 cpu MHz : 1399.790 cache size : 256 KB fdiv_bug: no hlt_bug : no f00f_bug: no coma_bug: no fpu : yes fpu_exception : yes cpuid level : 1 wp : yes flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr sse syscall mmxext 3dnowext 3dnow bogomips: 2791.83 Will this be included in hardrake, as Known bugs : (pass mem=nopentium instead of none) Stef
Re: [Cooker] mem=nopentium and AMD
Here - Here! ... differences between AMD and Intel based machines. Even th GCC compiler has AMD specific compiling optimizations. Some have argued that we should also have an AMD optimized kernel . . . . ... _ Chat with friends online, try MSN Messenger: http://messenger.msn.com
Re: [Cooker] mem=nopentium and AMD
An athlon/duron processor optimized kernel would be good (spec, perfs) ! Is the installer can detect and install it ? From: Mike Calloway [EMAIL PROTECTED] Reply-To: [EMAIL PROTECTED] To: [EMAIL PROTECTED] Subject: Re: [Cooker] mem=nopentium and AMD Date: Wed, 27 Feb 2002 07:20:03 -0500 Here - Here! ... differences between AMD and Intel based machines. Even th GCC compiler has AMD specific compiling optimizations. Some have argued that we should also have an AMD optimized kernel . . . . ... _ Chat with friends online, try MSN Messenger: http://messenger.msn.com = _ Téléchargez MSN Explorer gratuitement à l'adresse http://explorer.msn.fr/intl.asp.
Re: [Cooker] mem=nopentium and AMD
On Wed, 27 Feb 2002, Stéphane Teletchéa wrote: Is it realted only to this CPU's (from where do you take that values), i'm seeing it and here is what provies cat /proc/cpuifo : processor : 0 vendor_id : AuthenticAMD cpu family : 6 model : 6 model name : AMD Athlon(tm) XP 1600+ stepping: 2 Funny, I have an XP 1600+ as well, and the bug never happens for me. But notice the following: Crashes with 3d apps + AGP are not neccesarily related to this specific bug? It can also be your chipset's agp implementation or even the videocard +drivers (dri/nvidia) you use. Is your problem _really_ consistently fixed by using nopentium? Will this be included in hardrake, as Known bugs : (pass mem=nopentium instead of none) And what if I do not use AGP? the bug will not happen and I do not need this option. Danny Stef
Re: [Cooker] mem=nopentium and AMD
On Wed Feb 27 12:52 +0100, Stéphane Teletchéa wrote: Is it realted only to this CPU's (from where do you take that values), i'm seeing it and here is what provies cat /proc/cpuifo : processor : 0 vendor_id : AuthenticAMD cpu family : 6 model : 6 model name : AMD Athlon(tm) XP 1600+ stepping: 2 cpu MHz : 1399.790 cache size : 256 KB fdiv_bug: no hlt_bug : no f00f_bug: no coma_bug: no fpu : yes fpu_exception : yes cpuid level : 1 wp : yes flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr sse syscall mmxext 3dnowext 3dnow bogomips: 2791.83 I've been running 2.4 on my Duron 650 with great stability without the mem=nopentium option. Perhaps a doc should be placed on the mandrake site explaining the general issue and that it may not hurt to remove the option, but at the first sign of stability problems, mem=nopentium should be restored. -- Levi Ramsey [EMAIL PROTECTED] [EMAIL PROTECTED] When it comes down to desperation, You make the best of your situation. Linux 2.4.17-20mdk 1:01am up 1 day, 10:43, 13 users, load average: 0.25, 0.18, 0.16 msg57247/pgp0.pgp Description: PGP signature
Re: [Cooker] mem=nopentium and AMD
On Wed Feb 27 12:38 +, Cosmic Flo wrote: An athlon/duron processor optimized kernel would be good (spec, perfs) ! Is the installer can detect and install it ? From: Mike Calloway [EMAIL PROTECTED] Reply-To: [EMAIL PROTECTED] To: [EMAIL PROTECTED] Subject: Re: [Cooker] mem=nopentium and AMD Date: Wed, 27 Feb 2002 07:20:03 -0500 Here - Here! ... differences between AMD and Intel based machines. Even th GCC compiler has AMD specific compiling optimizations. Some have argued that we should also have an AMD optimized kernel . . . . I'm running a K7 kernel, and I'm not noticing anything different. Now if X, qt, gtk were K7-optimized, that would be a different story, I expect. Most of the K7-optimizations, as i recall, are floating-point, which is avoided in the kernel. -- Levi Ramsey [EMAIL PROTECTED] [EMAIL PROTECTED] When it comes down to desperation, You make the best of your situation. Linux 2.4.17-20mdk 1:01am up 1 day, 10:43, 13 users, load average: 0.25, 0.18, 0.16 msg57248/pgp0.pgp Description: PGP signature
Re: [Cooker] mem=nopentium and AMD
Le Mercredi 27 Février 2002 13:22, [EMAIL PROTECTED] a écrit : On Wed, 27 Feb 2002, Stéphane Teletchéa wrote: Is it realted only to this CPU's (from where do you take that values), i'm seeing it and here is what provies cat /proc/cpuifo : processor : 0 vendor_id : AuthenticAMD cpu family : 6 model : 6 model name : AMD Athlon(tm) XP 1600+ stepping: 2 Funny, I have an XP 1600+ as well, and the bug never happens for me. But notice the following: Crashes with 3d apps + AGP are not neccesarily related to this specific bug? It can also be your chipset's agp implementation or even the videocard +drivers (dri/nvidia) you use. Is your problem _really_ consistently fixed by using nopentium? I never checked this option but before having to do so (not now, i'm working on this machine !!), i would like to know how to trace it !!! Stef Will this be included in hardrake, as Known bugs : (pass mem=nopentium instead of none) And what if I do not use AGP? the bug will not happen and I do not need this option. May be another way, but nvidia'agp module is loaded, so i assume that they are reliable on themselves !) Danny Stef
Re: [Cooker] mem=nopentium and AMD
On Wed, 27 Feb 2002, Stéphane Teletchéa wrote: I never checked this option but before having to do so (not now, i'm working on this machine !!), i would like to know how to trace it !!! Stef This is not so easy, and part of the reason I think there is no fix yet. Only, if you have crashes when trying 3D apps which magically go away when you use the nopentium stuff it is likely you are affected. As far as I know the problem started in with the 750 Mhz series, but I could be completly wrong. (cannot find any good info about it). May be another way, but nvidia'agp module is loaded, so i assume that they are reliable on themselves !) huh? I do not understand what you mean. The problem (IIRC) is that memory allocated for AGP is also alocated for other stuff. This causes memory to be corrupted. If you are not using an AGP driver, nothing gets doubly allocated and there is no problem. As I understand it now, kernel developers are trying to find out a different way to see if the memory is mapped already, this would fix the issue in a good way. Danny
Re: [Cooker] mem=nopentium and AMD
Le Mardi 26 Février 2002 13:08, Robert Fox a écrit : I just performed an upgrade from a few week old Cooker to the most recent (kernel-2.4.17-21mdk) on my Asus A7V based Athelon TB machine. After the upgrade - the lilo.conf file is again modified and my manual entries of mem=nopentium were removed. My machine has a nasty tendancy to hard crash when I play BZFLAG for extended periods without the mem=nopentium option. I also read somewhere that this options if highly recommended (for stability) for AMD based mahines. Why doesn't the DrakX install catch this and do this automatically? Thx, R.Fox Yu're right, i've been told of this option (but couldn't remember the exact spelling) by an AMD seller at the Paris' Linux expo a month ago. May be useful to have it directly in lilo, as they recommend it. Any objection ? Stef
Re: [Cooker] mem=nopentium and AMD
[EMAIL PROTECTED] (Robert Fox) writes: [...] My machine has a nasty tendancy to hard crash when I play BZFLAG for extended periods without the mem=nopentium option. I also read somewhere that this options if highly recommended (for stability) for AMD based mahines. Why doesn't the DrakX install catch this and do this automatically? because I really think it should be fixed in the kernel? if mem=nopentium is *needed*, why not have it by default!
Re: [Cooker] mem=nopentium and AMD
On Tue, 26 Feb 2002, Robert Fox wrote: Bottom line is - my machine crashes HARD without this option - and I would venture to guess there are several others out there who also have AMD based machines - and some of these people may not be technical enough to find and/or install this option in lilo.conf - too hard for a newbie. True. This is a workaround for the cache consistency problems that occur with AGP video cards and AMD CPUs. It hurts performance, so it shouldn't be used everywhere. At the same time, I haven't heard of a real fix (which is non-trivial) for the problem, so waiting for one in the kernel may not make sense either. --- Stan Seibert
Re: [Cooker] mem=nopentium and AMD
[EMAIL PROTECTED] (Robert Fox) writes: [...] My machine has a nasty tendancy to hard crash when I play BZFLAG for extended periods without the mem=nopentium option. I also read somewhere that this options if highly recommended (for stability) for AMD based mahines. Why doesn't the DrakX install catch this and do this automatically? because I really think it should be fixed in the kernel? if mem=nopentium is *needed*, why not have it by default! [...] Agreed it should be fixed in the kernel - but how long can we afford to wait - and what are the consequences of NOT adding the option for the next (imminent) Mandrake release?? patching the kernel to have this is easy. I don't care if the standard kernel doesn't have it. AFAIK they (kernel people) say it's not a clean solution, nor does it fix everything. As for me, I don't care wether it's ugly or not. Having it lilo.conf means it will stay there after upgrading to fixed kernel. The right place to fix is in the kernel.
Re: [Cooker] mem=nopentium and AMD
Robert Fox wrote: Agreed it should be fixed in the kernel - but how long can we afford to wait - and what are the consequences of NOT adding the option for the next (imminent) Mandrake release?? Actually, there seems to be a bug in drakboot, where there is mem=nonpentium (By default? why?), but when you go to click OK, it says that mem must be a number. Obviously, it doesn't have to be a number. On the other hand, I have an A7V motherboard and have never had a stability problem that I could attribute to this option. So, forcing this on all AMD users is stupid, and shouldn't be done. What should be done is the boot configuration should be fixed, and it could also provide a description of the nopentium flag. -- Sincerely, David Walluck [EMAIL PROTECTED]
Re: [Cooker] mem=nopentium and AMD
On Tue, 2002-02-26 at 18:47, David Walluck wrote: On the other hand, I have an A7V motherboard and have never had a stability problem that I could attribute to this option. So, forcing this on all AMD users is stupid, and shouldn't be done. What should be done is the boot configuration should be fixed, and it could also provide a description of the nopentium flag. -- You have probably not had this stability problem because you don't have an NVidia based graphic card (I have a GForce2 DDR) and use the NVidia drivers. Still, even AMD people have recognized this problem formally and recommend the use of this parameter . . . There is no forcing this on all users and it's not very stupid. It's also possible you don't stress your machine enough to see the probem. I agree about the description for the flag - but there should at least be an option during install to add it or not . . . Cheers, R. Fox
Re: [Cooker] mem=nopentium and AMD
Robert Fox wrote: You have probably not had this stability problem because you don't have an NVidia based graphic card (I have a GForce2 DDR) and use the NVidia drivers. Still, even AMD people have recognized this problem formally and recommend the use of this parameter . . . Actually, I have this very card, and the NVIDIA drivers, though I get some kind of AGP error (before the server even starts). Mandrake can't be held responsible for *commercial* drivers though (What's with the kernel complaining about the licenses though? Is it illegal to have non-GPL modules? I think this is a patch that was added to the mdk kernel.). Anyway, I like the speed increase of the NVIDIA drivers, but the stability should come first. I agree about the description for the flag - but there should at least be an option during install to add it or not . . . It *is* an option. Hopefully, someone else on the list is listening and will fix the problem with drakboot. -- Sincerely, David Walluck [EMAIL PROTECTED]
Re: [Cooker] mem=nopentium and AMD
On Tuesday 26 Feb 2002 13:43, Pixel wrote: [EMAIL PROTECTED] (Robert Fox) writes: [...] My machine has a nasty tendancy to hard crash when I play BZFLAG for extended periods without the mem=nopentium option. I also read somewhere that this options if highly recommended (for stability) for AMD based mahines. Why doesn't the DrakX install catch this and do this automatically? because I really think it should be fixed in the kernel? if mem=nopentium is *needed*, why not have it by default! What? The provided Mandrake kernel is for all types of processors, not just AMD. You wouldn't want mem=nopentium is you have a pentium. So command line option is the only option as far as I can see. -- Peter Ruskin, Wrexham, Wales. AMD Athlon XP 1600+, 512MB RAM. Registered Linux User 219434. Mandrake Linux release 8.1 (Vitamin) Kernel 2.4.8-34.1mdk-win4lin, XFree86 4.1.0, patch level 21mdk. KDE: 2.2.2. Qt: 2.3.2. Up 2 days 8 minutes.
Re: [Cooker] mem=nopentium and AMD
On Tue, 2002-02-26 at 15:01, Peter Ruskin wrote: On Tuesday 26 Feb 2002 13:43, Pixel wrote: [EMAIL PROTECTED] (Robert Fox) writes: [...] My machine has a nasty tendancy to hard crash when I play BZFLAG for extended periods without the mem=nopentium option. I also read somewhere that this options if highly recommended (for stability) for AMD based mahines. Why doesn't the DrakX install catch this and do this automatically? because I really think it should be fixed in the kernel? if mem=nopentium is *needed*, why not have it by default! What? The provided Mandrake kernel is for all types of processors, not just AMD. You wouldn't want mem=nopentium is you have a pentium. So command line option is the only option as far as I can see. -- Peter Ruskin, Wrexham, Wales. AMD Athlon XP 1600+, 512MB RAM. Registered Linux User 219434. Mandrake Linux release 8.1 (Vitamin) Kernel 2.4.8-34.1mdk-win4lin, XFree86 4.1.0, patch level 21mdk. KDE: 2.2.2. Qt: 2.3.2. Up 2 days 8 minutes. The installer could check for the kind of CPU and adjust in consequence...
Re: [Cooker] mem=nopentium and AMD
I never implied that the this option should be used for pentium based machines. During the install, it is possible to determine whether the processor is AMD or Intel based - and IF AMD - then add 'option' Just do this: cat /proc/cpuinfo and look at the Vendor_ID and model name fields. This is possible. Cheers, R.Fox On Tue, 2002-02-26 at 21:01, Peter Ruskin wrote: On Tuesday 26 Feb 2002 13:43, Pixel wrote: [EMAIL PROTECTED] (Robert Fox) writes: [...] My machine has a nasty tendancy to hard crash when I play BZFLAG for extended periods without the mem=nopentium option. I also read somewhere that this options if highly recommended (for stability) for AMD based mahines. Why doesn't the DrakX install catch this and do this automatically? because I really think it should be fixed in the kernel? if mem=nopentium is *needed*, why not have it by default! What? The provided Mandrake kernel is for all types of processors, not just AMD. You wouldn't want mem=nopentium is you have a pentium. So command line option is the only option as far as I can see. -- Peter Ruskin, Wrexham, Wales. AMD Athlon XP 1600+, 512MB RAM. Registered Linux User 219434. Mandrake Linux release 8.1 (Vitamin) Kernel 2.4.8-34.1mdk-win4lin, XFree86 4.1.0, patch level 21mdk. KDE: 2.2.2. Qt: 2.3.2. Up 2 days 8 minutes.
Re: [Cooker] mem=nopentium and AMD
David Walluck [EMAIL PROTECTED] writes: It *is* an option. Hopefully, someone else on the list is listening and will fix the problem with drakboot. I will (try)
Re: [Cooker] mem=nopentium and AMD
Peter Ruskin [EMAIL PROTECTED] writes: My machine has a nasty tendancy to hard crash when I play BZFLAG for extended periods without the mem=nopentium option. I also read somewhere that this options if highly recommended (for stability) for AMD based mahines. Why doesn't the DrakX install catch this and do this automatically? because I really think it should be fixed in the kernel? if mem=nopentium is *needed*, why not have it by default! What? The provided Mandrake kernel is for all types of processors, not just AMD. You wouldn't want mem=nopentium is you have a pentium. So command line option is the only option as far as I can see. i meant of course for Athlons, not for any CPUs. if (is_on_athlon_or_duron) { do_the_same_thing_as_mem_no_pentium_is_doing(); } But anyway, our kernel people tells it's ugly/non-necessary. So I won't bother with this anymore. If you have some pb with Athlons, tell [EMAIL PROTECTED], and don't tell me to add mem=nopentium ;p
Re: [Cooker] mem=nopentium and AMD
On Tue, 2002-02-26 at 12:47, David Walluck wrote: Robert Fox wrote: Agreed it should be fixed in the kernel - but how long can we afford to wait - and what are the consequences of NOT adding the option for the next (imminent) Mandrake release?? Actually, there seems to be a bug in drakboot, where there is mem=nonpentium (By default? why?), but when you go to click OK, it says that mem must be a number. Obviously, it doesn't have to be a number. nopentium was probably written there because you performed an upgrade from a system that already had the option appended to LILO... I think it would be very important to add this since it affects the majority of AMD cpus used with the AGP port... The change that mem=nopentium does is simply telling the kernel to use 4k pages instead of 4M pages and almost no programs can use 4M pages anyway so the performance loss is absolutely imperceptible. Many person reported big stability problems when 8.1 came out and this option can prevent that... For those interested in knowing if their CPU is affected, you can make cat /proc/cpuinfo and if the tree following lines are not identical then your CPU probably has the bug: cpu family : 6 model : 2 stepping: 2 Regards, Baal
Re: [Cooker] mem=nopentium and AMD
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 As few days ago I read a discussion on this (possibly on lkml), anyway, patching the kernel for this might not be so easy as: - - - Do you know exactly which chips are affected? My Athlon XP sure is not, and I would NOT like this option since it decrease performance drastically. And for real pentium processors it probably causes the kernel not to boot (didn't try though because i do not have a Pentium). -- there is not simple test yet to determine whether your processor is affected. - - - This option only makes the bug less likely to occur, it does not fix the issue. - - - If no agpgart(or nvidia agp) is used, the bug does not occur and the option only serves to degrade performance (think servers here). Danny On Tuesday 26 February 2002 18:47, you wrote: Robert Fox wrote: Agreed it should be fixed in the kernel - but how long can we afford to wait - and what are the consequences of NOT adding the option for the next (imminent) Mandrake release?? Actually, there seems to be a bug in drakboot, where there is mem=nonpentium (By default? why?), but when you go to click OK, it says that mem must be a number. Obviously, it doesn't have to be a number. On the other hand, I have an A7V motherboard and have never had a stability problem that I could attribute to this option. So, forcing this on all AMD users is stupid, and shouldn't be done. What should be done is the boot configuration should be fixed, and it could also provide a description of the nopentium flag. - - -- `...we might as well start with where your hand is now.' Arthur said, `So which way do I go?' `Down,' said Fenchurch, `on this occaision.' He moved his hand. `Down,' she said, `is in fact the other way.' `Oh yes.' - - - Arthur trying to discover which part of Fenchurch is wrong. -BEGIN PGP SIGNATURE- Version: GnuPG v1.0.6 (GNU/Linux) Comment: For info see http://www.gnupg.org iD8DBQE8e/27aeiN+EU2vEIRAslhAJ0e02waEhvZzzmg3kD/QcWki2QJsgCfZSsx +DwAJTLoe49tBsEYdwHFMjQ= =P1nJ -END PGP SIGNATURE-