Re: Alpha SMP status
* James Andrewartha wrote: Thanks again to the alpha maintainers, and extra thanks if you can sneak a fix for bug 402253 in before etch (I even included a patch ;-). Patch added, thanks. Norbert -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Alpha SMP status
On Wed, 13 Dec 2006, Aneesh Kumar K.V wrote: May be it is running out of space for per cpu area. You can try by increasing the PERCPU_ENOUGH_ROOM in include/linux/percpu.h and see if get pass. I looked other 64 bit archs x86_64 and ia64 and they both have different value. May be alpha also needs a different value. These patches have been useful for the AlphaCore-2 2.6.15-based kernels in getting SMP working, though we seem to also need to build part of the SCSI subsystem into the kernel as well. Instead of increasing it to another constant, How about this http://marc.theaimsgroup.com/?l=linux-kernelm=116593521703170w=2 If you agree Can you send an ACK to akpm on this ? This does fix the No per-cpu room for modules error, but we then go back to the module ipv6: Relocation overflow vs section 27 error, as seen in http://archives.neohapsis.com/archives/linux/axp/2005-q1/0157.html and as early as 2.5.72: http://lkml.org/lkml/2003/6/24/64 There's a patch at http://lkml.org/lkml/2004/1/11/42, but that's already been applied. http://lkml.org/lkml/2004/1/8/99 suggests it might be a gcc issue - I'm using gcc (GCC) 4.1.2 20061028 (prerelease) (Debian 4.1.1-19). -- # TRS-80 trs80(a)ucc.gu.uwa.edu.au #/ Otherwise Bub here will do \ # UCC Wheel Member http://trs80.ucc.asn.au/ #| what squirrels do best | [ There's nobody getting rich writing ]| -- Collect and hide your | [ software that I know of -- Bill Gates, 1980 ]\ nuts. -- Acid Reflux #231 / -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Alpha SMP status
On Tue, Dec 12, 2006 at 11:26:18AM +0530, Aneesh Kumar K.V wrote: On Tue, 2006-12-12 at 10:23 +0900, James Andrewartha wrote: Hi all, Some of you may be aware that SMP has been broken on Alpha for most of 2.6. As of linux-image-2.6.18-3-smp, I'm pleased to report that it boots and runs acceptably because scsi_mod is compiled in (bug 369517). However, things like ip_conntrack and ipv6 are still broken - inserting them gives errors like Could not allocate 8 bytes percpu data. This explains why all the scsi drivers were broken - scsi_mod also couldn't allocate percpu data and so wasn't loaded. As with scsi, you can work around this by compiling them (ipv6 etc.) in to the kernel rather than as modules. Anyway, this points to the core of the problem as being something to do with percpu data. Hopefully this narrows it down enough for someone familiar with the kernel to have a go at finding and fixing the problem. May be it is running out of space for per cpu area. You can try by increasing the PERCPU_ENOUGH_ROOM in include/linux/percpu.h and see if get pass. I looked other 64 bit archs x86_64 and ia64 and they both have different value. May be alpha also needs a different value. These patches have been useful for the AlphaCore-2 2.6.15-based kernels in getting SMP working, though we seem to also need to build part of the SCSI subsystem into the kernel as well. --Jay++ --- Jay A EstabrookHPTC - XC I B Hewlett-Packard Company - ZKO1-3/D-B.8 (603) 884-0301 110 Spit Brook Road, Nashua NH 03062 [EMAIL PROTECTED] --- --- linux-2.6.15/include/asm-alpha/percpu.h.orig2006-01-03 06:21:10.0 +0300 +++ linux-2.6.15/include/asm-alpha/percpu.h 2006-03-22 16:46:05.0 +0300 @@ -1,6 +1,8 @@ #ifndef __ALPHA_PERCPU_H #define __ALPHA_PERCPU_H +#definePERCPU_ENOUGH_ROOM (192*1024) + #include asm-generic/percpu.h #endif /* __ALPHA_PERCPU_H */ --- linux-2.6.15/kernel/module.c.ORIG 2006-02-28 17:09:53.0 -0500 +++ linux-2.6.15/kernel/module.c2006-02-28 17:11:57.0 -0500 @@ -293,8 +293,8 @@ return ptr; } - printk(KERN_WARNING Could not allocate %lu bytes percpu data\n, - size); + printk(KERN_WARNING Could not allocate %lu bytes percpu data for module %s\n, + size, name); return NULL; }
Re: Alpha SMP status
On Tue, 2006-12-12 at 10:55 -0500, Jay Estabrook wrote: On Tue, Dec 12, 2006 at 11:26:18AM +0530, Aneesh Kumar K.V wrote: On Tue, 2006-12-12 at 10:23 +0900, James Andrewartha wrote: Hi all, Some of you may be aware that SMP has been broken on Alpha for most of 2.6. As of linux-image-2.6.18-3-smp, I'm pleased to report that it boots and runs acceptably because scsi_mod is compiled in (bug 369517). However, things like ip_conntrack and ipv6 are still broken - inserting them gives errors like Could not allocate 8 bytes percpu data. This explains why all the scsi drivers were broken - scsi_mod also couldn't allocate percpu data and so wasn't loaded. As with scsi, you can work around this by compiling them (ipv6 etc.) in to the kernel rather than as modules. Anyway, this points to the core of the problem as being something to do with percpu data. Hopefully this narrows it down enough for someone familiar with the kernel to have a go at finding and fixing the problem. May be it is running out of space for per cpu area. You can try by increasing the PERCPU_ENOUGH_ROOM in include/linux/percpu.h and see if get pass. I looked other 64 bit archs x86_64 and ia64 and they both have different value. May be alpha also needs a different value. These patches have been useful for the AlphaCore-2 2.6.15-based kernels in getting SMP working, though we seem to also need to build part of the SCSI subsystem into the kernel as well. Instead of increasing it to another constant, How about this http://marc.theaimsgroup.com/?l=linux-kernelm=116593521703170w=2 If you agree Can you send an ACK to akpm on this ? -aneesh -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Alpha SMP status
Hi all, Some of you may be aware that SMP has been broken on Alpha for most of 2.6. As of linux-image-2.6.18-3-smp, I'm pleased to report that it boots and runs acceptably because scsi_mod is compiled in (bug 369517). However, things like ip_conntrack and ipv6 are still broken - inserting them gives errors like Could not allocate 8 bytes percpu data. This explains why all the scsi drivers were broken - scsi_mod also couldn't allocate percpu data and so wasn't loaded. As with scsi, you can work around this by compiling them (ipv6 etc.) in to the kernel rather than as modules. Anyway, this points to the core of the problem as being something to do with percpu data. Hopefully this narrows it down enough for someone familiar with the kernel to have a go at finding and fixing the problem. Thanks again to the alpha maintainers, and extra thanks if you can sneak a fix for bug 402253 in before etch (I even included a patch ;-). -- # TRS-80 trs80(a)ucc.gu.uwa.edu.au #/ Otherwise Bub here will do \ # UCC Wheel Member http://trs80.ucc.asn.au/ #| what squirrels do best | [ There's nobody getting rich writing ]| -- Collect and hide your | [ software that I know of -- Bill Gates, 1980 ]\ nuts. -- Acid Reflux #231 / -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Alpha SMP status
On Tue, 2006-12-12 at 10:23 +0900, James Andrewartha wrote: Hi all, Some of you may be aware that SMP has been broken on Alpha for most of 2.6. As of linux-image-2.6.18-3-smp, I'm pleased to report that it boots and runs acceptably because scsi_mod is compiled in (bug 369517). However, things like ip_conntrack and ipv6 are still broken - inserting them gives errors like Could not allocate 8 bytes percpu data. This explains why all the scsi drivers were broken - scsi_mod also couldn't allocate percpu data and so wasn't loaded. As with scsi, you can work around this by compiling them (ipv6 etc.) in to the kernel rather than as modules. Anyway, this points to the core of the problem as being something to do with percpu data. Hopefully this narrows it down enough for someone familiar with the kernel to have a go at finding and fixing the problem. May be it is running out of space for per cpu area. You can try by increasing the PERCPU_ENOUGH_ROOM in include/linux/percpu.h and see if get pass. I looked other 64 bit archs x86_64 and ia64 and they both have different value. May be alpha also needs a different value. -aneesh -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]