Re: G5 blade center under debian
Hi, > I've just updated the installer image to have no POWER4+ optimization, > but the language chooser seems to be broken. I've no idea why. Nice, it gets a lot further on POWER3 now :) It stops right here: time_init: decrementer frequency = 93.748235 MHz time_init: processor frequency = 375.00 MHz I wonder if serial and serial console is disabled in this kernel... It also stops at the same point on a JS20 blade: time_init: decrementer frequency = 199.839239 MHz time_init: processor frequency = 1600.00 MHz Is hvc console disabled too? :) Anton -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: number of Power 5 processors supported?
Hi, > Does anyone know how many processors are supported on Power 5 SMP machines > with Linux? > > I'm hoping to buy a 16 processor system and am cureous if theres any hope > of Linux being able to use all 16 procs. Yes. We regularly boot POWER5 machines with 128 threads (64 way). Note that debian wont install out of the box but there is work ongoing to fix that. Anton -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: PPC64 kernel build toolchain?
> >Can anyone tell me where to get a toolchain that can build a PPC64 (NOT > >PPC32) kernel? I've now got a few PPC64 machines running Debian that > >I'd like to run an actual 64-bit kernel on... > > Maybe you'll have better luck. I built one with crosstool-031 which was > very easy. Unfortunately when I try compiling a ppc 64 kernel 2.6.11.6 it > stops with: It looks like you are picking up the host compiler and not the ppc32 compiler. Check out arch/ppc64/boot/Makefile and the CROSS32_COMPILE define. Anton -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: PPC64 kernel build toolchain?
> Mucking with the environment variables as you suggest fixed my problem. > Thanks! I have now compiled a PPC 64 kernel, but sadly do not have a G5 > to try it on. :( Unfortunately I cant help you there :) > I noticed gcc uses the -m64 option when compiling the 64 bit things. When > no -mXY is given does gcc default to -m32? I suppose I could find the > answer in the docs somewhere. With a biarch toolchain you have the choice of the default when configuring gcc. To make the default 32bit, use: --with-cpu=default32 Anton -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
POWER3 install CD
Hi, I was looking for a POWER3 debian install CD and came across the installer ports status page: http://www.debian.org/devel/debian-installer/ports-status It mentions these as working: powerpc IBM RS6000 power3 IBM RS6000 power4 IBM RS6000 It turns out these targets overlap with the next two: IBM pSeries IBM iseries Which are marked as unavailable. I also tried booting the CD image on a POWER3 box and it did not work. What options are used to build the ISO? Anton -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: ppc64 pseries 2.6.11-2 .deb kernel available
> just wondering, is the timebase suppose to look like this? and where are > the bogomips gone? Yep, the timebase is externally clocked on those machines. And since bogomips is useless there is no need to advertise it here :) Anton -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: reiserfs empirical study (very long)
> The big endian patches change the code to use little endian ordering > for all on-disk structures. IMO this is a mistake, and certainly > costs a dear performance penalty, because on big endian processors, > this method requires converting endianness both ways (reading and > writing) for all meta data. Sane architectures (sparc64, ppc) have load/store with byte swap instructions and if reiserfs is using them you shouldnt see a performance penalty. cpu_to_le* etc make use of them. Anton
Re: XFree86 4.1.0: call for help
Hi,
> - r128 doesn't work without UseFBDev at least on Pismos (I suspect all
> UniNorth Macs) because we're now trying to do ISA I/O with the help of a new
> system call, but the server crashes. It works fine on those machines without
> it, so what would we lose by disabling it in r128 #ifdef __powerpc__ ? Or what
> would be a better way to determine if it should be used or not?
What system call is this?
Do you know if the patch by davem to make X mmap("/proc/bus/pci/...") has
made it into 4.1.0? It makes life a lot easier for people with non transparent
PCI buses (at the moment I hardwire the offsets in for my RS6000s).
Anton
Re: XFree86 4.1.0: call for help
> btw mmap("/proc/bus/pci/...") was added around 2.4.4 right ?
Yep 2.4.5. From memory the X patches I saw did not allow the
X server to fall back to mmap("/dev/mem") which is important.
Anton
Re: reiserfs empirical study (very long)
Hi, > even a small start would be helping me get better RS/6000 support in > ybin, since the boot-floppies completely rely on ybin to make > yaboot systems bootable. I have debian running on a number of RS6000s. Ignoring the install, all that is required (on POWER3 machines) is to fix libc to understand the large cacheline size. At the moment I am unpacking the deb, binary editing it and repacking it (a trick paulus showed me). Longer term Benh has patches to export the cacheline size from the kernel to libc which will fix the problem. Hopefully we can get something into the boot-floppies libc to handle this (alternatively just get rid of the dcbf in memset(, 0, 0)) Anton
Re: PowerPC 64-bit kernels
> So, as I've noticed that I can't successfully compile a 64bit powerpc > kernel using any sort of 'normal' kernels (or it may be a GCC issue). > Precisely what toolchain/kernel sources (ie give me URL's or debian > package names) do I need to build a ppc64 kernel? Can you elaborate? 2.6 should build and boot right out of the box. Where did you get your 64bit toolchain from? Anton
debian-installer on ppc64
Hi, We tested debian-installer (daily build from a day ago) on a ppc64 box and found a few issues. To keep life interesting, we also used a 2.6 kernel. 1. Installer requires devfs support. In 2.6 devfs is pretty much deprecated, so it would be nice if d-i could handle a non devfs kernel. It doesnt look like there are too many places that care. The initrd had an empty /dev, thats easily fixed. It looks like the stuff that parses /proc/partitions expects devfs style names but I havent looked into the code yet. 2. Installer cant find any packages. Hacking the kernel to return "ppc" instead of "ppc64" fixes this problem. Im guessing the sparc64 port of d-i has solved this problem, we just need it too. 3. No fdisk found. For some reason only mac-fdisk was available in the install image. 4. shm is not a valid filesystem in 2.6. Changing /sbin/init to use tmpfs fixed this. Anton
Re: debian-installer on ppc64
Hi,
> Mmm, maybe it should be possible to create a proper kernel package for
> your boxes, could you send me a .config file for a 2.4.22 kernel ? Will
> the standard linuxppc_2_4 tree do, or do you need special patches ?
We need to get the 64bit ppc cross compiler into debian, it sounded
like there are a few people working on this.
> Well, this would need the corresponding magic to be added arch-detect. I
> guess ppc64 uses ppc userland, right ? Could you provide us with the
> output of lspci ? Or maybe hack arch-detect yourself ?
Yes, almost all of our userspace is shared with ppc32. archdetect looks
to be doing the correct thing, on my ppc64 box it reports:
powerpc/chrp
>From a quick look around one problem seems to be the output of
udpkg --print-architecture. How does this (untested as of yet)
patch look?
diff -u -r1.44 udpkg.c
--- tools/udpkg/udpkg.c 2 Nov 2003 15:55:30 - 1.44
+++ tools/udpkg/udpkg.c 15 Nov 2003 06:29:26 -
@@ -61,6 +61,7 @@
{ "parisc", "hppa" },
{ "parisc64", "hppa" },
{ "ppc", "powerpc"},
+{ "ppc64", "powerpc"},
};
if (uname(&name) < 0)
> Same as above, d-i doesn't support subarches yet. That said, parted
> should be available, and normally you should be able to use partitioner
> and/or autopartkit transparently. What partitioning scheme are your box
> using ?
It uses standard DOS partitions. I usually use fdisk to create them.
> Mmm, no idea about this one, maybe a special 2.6 kernel support for d-i
> would be nice to have, a bit like woody's boot-floppies supported 2.2
> and 2.4 kernels ?
As Christoph mentioned, calling it tmpfs should work across 2.4 and 2.6.
Anton
Re: ibm compiler for 64bit ppc
Hi, > I have downloaded xlc ibm compiler for suse (trial version) but ppc64 > isn't supported by alien. So I unpacked everythings but I didn't > success in running it (xlc doesn't found 1501-231 message). Does > somebody already have done it ? > > I really want to run it under debian since I don't want to switch to > suse for our project (mix of blade JS20 center,OpenPower,xserve...). You need the install scripts in the rpm. Luckily alien handles the conversion OK. I got around the ppc64 target issue by unpacking all of the rpms and changing the target from ppc64 to powerpc: mkdir tmp cd tmp for i in ../*rpm do alien -g --scripts $i done for i in `find . -name control` do perl -i -p -e 's/ppc64/powerpc/' $i done for i in `ls -d * |grep -v orig` do cd $i ./debian/rules binary cd .. done Then install the debs and run the configure script. You can try and convert /opt/ibmcmp/vac/7.0/bin/new_install to understand debs or else just run the vac_configure script directly. Assuming you dont have a biarch toolchain (which is the case with debian), and you have a 64bit capable toolchain in usr/local/ppc64, you can run: /opt/ibmcmp/vac/7.0/bin/vac_configure -smprt /opt/ibmcmp/xlsmp/1.5 -mass /opt/ibmcmp/xlmass/4.1 -vac /opt/ibmcmp/vac/7.0 -install /opt/ibmcmp/vac/7.0/etc/vac.base.cfg -gcc /usr -gcc64 /usr/local/ppc64 -vacpp /opt/ibmcmp/vacpp/7.0 -vacpprt /opt/ibmcmp/vacpp/7.0 -vaclic /opt/ibmcmp/vac/7.0 And for xlf: /opt/ibmcmp/xlf/9.1/bin/xlf_configure -smprt /opt/ibmcmp/xlsmp/1.5 -mass /opt/ibmcmp/xlmass/4.1 -xlf /opt/ibmcmp/xlf/9.1 -install /opt/ibmcmp/xlf/9.1/etc/xlf.base.cfg -gcc /usr -gcc64 /usr/local/ppc64 -xlfrt /opt/ibmcmp/xlf/9.1 -xlflic /opt/ibmcmp/xlf/9.1 PS we really need a biarch toolchain but are blocked by the debian release cycle (we need an updated glibc). Anton
Re: Fwd: SCSI-disk
> It sounds a lot like the cases in i386 where a given hardware module > isn't compiled into the kernel. It does. Is the SYM53C8XX driver compiled into the boot disk kernel? Anton -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
strace issues
Hi, I submitted some patches for strace on ppc a few months ago but nothing has happened. The worst one is the fact that strace tries to insert an intel assembly instruction in order to lock a child process in a busy loop. This makes the -f option completely unreliable. Can someone do an NMU for this? Id like to get a reliable strace back. I cant find my bug report (do bug reports ever disappear?) but someone else has reported it recently: #153750: strace: Bogus LOOP for PowerPC Also this will fix a problem with IPC syscalls: #144326: strace on ppc linux misreports IPC syscalls Anton
Re: Problem with a GXT3000P VGA Adapter in a RS6000 (44p 170)
> Does anybody have expirience with this Graphic-Adapter ? I can't even see > the Boot Console (over a serial Line it's ok). > When i use the Adapter, i see two small reflections of the console with > unreadble text. I think it is a syncronisation Problem. > I don't wanna use X in the moment, the text console would be enough. We can handle GXT2000 and GXT4000 using OF FB but the GXT3000 is quite weird, supporting it wont be simple. The GXT130P is basically a matrox G200, its supported by both OF FB and the accelerated matrox FB. We should put together a list of IBM cards that work (and dont work). Note GXT4000 support is not in OF FB yet, I have a patch from Paul which needs to be integrated. Anton
Re: IBM pSeries (p655) w/Power4 chipset?
Hi, > Is any one running Debian on an IBM pSeries server? > > I have a p655 on the way and I'm wondering what my chances are. > > I have fairly extensive experience with Debian, and no experience with > IBM Powerpc based machines or AIX. > > Any and all pointers and suggestions welcomed. Ive run Debian on pretty much everything from a p690 down to a B80. These days the only big issue is the kernel. I tend to tack a debian install root image onto a ppc64 kernel and tftp boot it. Unfortunately it whinges a bit (kernel version doesnt match the root image etc) but you can force it on. Im looking forwards to the new install system so I can build proper ppc64 install CDs. I tried to do it with the current boot floppies but got incredibly lost. Anton
Re: Debian on RS6K
Hi John, > Does debian support X on RS6K? > > I have IBM 256-bit graphics rasterizer [Fire GL1] card. I tried > using the generic vga settings but > it complains about the clocks line. Interesting, is this a standard RS6000 card? If so what is the IBM name for it? > Also, does debian support booting from hard disk now? It should work fine, although one of the guys here had some problems with yaboot on his 170. Anton
Re: tftp boot troubles (IBM p655)
> I have six shiny new IBM p655 nodes Great! > and I can't do a thing with them... Not great. > I'll save AIX aches for else where, but my Debian problem is that I > can't seem to get them to tftpboot (and they have no removable media). > > the get IP and the boot file name correctly from the DHCP server via > BOOTP, but they try and tftp from 255.255.255.255 rather than the > server I specify as "next-server" in the dhcp entry for the host or > the reasonable default of trying the BOOTP server's address. On my machines it will try and tftp from the bootp server address. If it doesnt respond then it falls back to the broadcast address. Ive never tried using next-server, although you might have some luck with: ibm,fw-clientipaddr ibm,fw-serveripaddr ibm,fw-gatewayipaddr ibm,fw-netmask Once it does boot there are two problems installing debian. Firstly we dont have an install image with a 64bit kernel. You can work around it by grabbing the ramdisk image and building a kernel with it. We would have ppc64 debian install images if the boot floppies stuff wasnt so hard to work out. Im hoping someone who knows what they are doing ends up building them :) Last problem is that the p655 contains a new SCSI chipset and IBM has decided to release a binary only linux driver. You will need to compile it in. Anton
Re: Testing Sarge on IBM pSeries systems
Hi Rolf, > How can I help out testing Sarge on IBM pSeries systems ( I have access to > a lab full of these systems )? Eventually Id like to see a debian CD that boots on pSeries boxes. I built a woody one recently but I run into problems during the install. Im not adventurous enough to rebuild the debian installer initrd and the kernel I tacked it onto didnt match - it got upset when it got to installing kernel/modules I think. Is anyone playing with the new installer on ppc? Is it worth trying yet? Anton
Re: New Kernel on a RS6000 44p 170
> When i understand you right, i have to install and compile the > Toolchain on my System and after that i can compile the > patched Kernel from penguinppc64 ? > Sorry, i'am not yet an Insider but where can > i find the 2.4.21-pre6 Code under Marcelo's Directory on kernel.org ? If it becomes all too hard you should be able to compile and run the 32bit kernel. We have a 270 and 170 that run it, if current 2.4 doesnt work then we should let Paul know :) Anton
Re: powerpc wmb()
Hi,
> wmb is defined as
>
> #define wmb() __asm__ __volatile__ ("eieio" : : : "memory")
>
> however, ``eieio'' has no effect on memory, that is
> Write Through Not Required, thus I think it should be
>
> #define wmb() __asm__ __volatile__ ("sync" : : : "memory")
While the original ppc spec did not guarantee eieio ordered writes
to caching allowed, write through not required, memory coherence
required storage, there have been modifications since:
http://www.rs6000.ibm.com/tech/ppc-chg2.html
Look for the change "Book II, Section 3.3, page 350"
My understanding is that the earlier chips completed stores in order
which allowed the above change to be made safely, but I am not 100%
on this.
Anton
Re: date problems
> I'm being freaked out by a really weird bug. > > iBook2, running testing, with 2.4.16-ben0 as the kernel. > > The machine is convinced it's a month ahead (as the headers on this mail > probably say). It sounds like a kernel bug that should be fixed in a more recent kernel. It wasnt a laptop but each time we booted the machine it jumped forward one month :) Anton
Re: debian on PPC IBM RS6000
> Dear Friends, > This is the first time that i write to you > and i don't know is this email destination > is rigth for my questions. > If No, please, tell me where email. > > I have: > IBM RISC System/6000 > model 43P-140 > other informations: > 640e 200 MHz > 7043 44-L0258 > firmware TIG 97112. > Actually, this PPC runs with AIX IBM, and > i would format all and install DEBIAN PPC. Good news is that Paulus has fixed one of the last problems with the 140 (firmware configures a stupidly large PCI window) The bad news is I dont think you can get a debian boot disk with the fixes. Is anyone looking at creating/testing debian boot disks for PREP? It should be easy to make a Power3 boot disk too (either 32bit or 64bit kernel will work) Anton
