Re: [OpenAFS] openafs w/ amd64 kernel and 32bit user space
In message [EMAIL PROTECTED],Sean O'Malley writes: I ran into the same problem =) I started hacking AFS to get userland stuff working, but put a dent in my head banging off the wall. I did manage to get it to compile, and started in with lint but that was about as far as I got. :) dont bother with lint. just collect the warnings. anything about implicit is likely bad news on 64-bit. after than, any casting to/from wrong size integers is possibly bad. afs has a habit of casting int's to void * and back again which is safe. that was the bulk of the work getting the ia64 client to run 64-bit. ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
Re: [OpenAFS] openafs w/ amd64 kernel and 32bit user space
On Tue, 27 Mar 2007, chas williams - CONTRACTOR wrote: dont bother with lint. just collect the warnings. anything about implicit is likely bad news on 64-bit. after than, any casting to/from wrong size integers is possibly bad. afs has a habit of casting int's to void * and back again which is safe. that was the bulk of the work getting the ia64 client to run 64-bit. Lint is just too slow at this point. only 4353 warnings, and only 2095 of them are implicit.. :) that is down from xmas break. =-) -- Sean O'Malley, Information Technologist Michigan State University - ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
Re: [OpenAFS] openafs w/ amd64 kernel and 32bit user space
On Fri, 23 Mar 2007, Carson Gaspar wrote: And sadly many user-land pieces still are not 64-bit clean. I know that I've had to fall back to 32-bit mode for several apps on my Solaris x86 server, as I didn't have the time energy to fix the crappy source. I ran into the same problem =) I started hacking AFS to get userland stuff working, but put a dent in my head banging off the wall. I did manage to get it to compile, and started in with lint but that was about as far as I got. :) It is one of those things where everyone if they get time should add a hack here and there to make it 64-bit safe when they see something. It isnt going to get done in a one weekend hack-a-thon until more is cleaned up. One of the bigger problems I ran into was because stuff is compiled more then once, you don't necessarily see the error right away. It gets lost in the rest of the compiler warnings that might be come from a different section of the compile. (just because it compiles cleanly doesnt mean it works and vice versa.) -- Sean O'Malley, Information Technologist Michigan State University - ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
Re: [OpenAFS] openafs w/ amd64 kernel and 32bit user space
Sean O'Malley wrote: On Fri, 23 Mar 2007, Carson Gaspar wrote: And sadly many user-land pieces still are not 64-bit clean. I know that I've had to fall back to 32-bit mode for several apps on my Solaris x86 server, as I didn't have the time energy to fix the crappy source. I ran into the same problem =) I started hacking AFS to get userland stuff working, but put a dent in my head banging off the wall. I did manage to get it to compile, and started in with lint but that was about as far as I got. :) To clarify, as 2 people have misunderstood my message, I have had _no_ problems with OpenAFS 64-bit (but I've only used it on RHEL4). I've had many problems with _other_ pieces of software, thus my sympathy for wanting a 32-bit user space. -- Carson ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
Re: [OpenAFS] openafs w/ amd64 kernel and 32bit user space
Chris, thanks for the confirmation that this would work. Since then I was able to put together a system as I described (Debian/sarge 32bit user space with a amd64 kernel and openafs) and I am pleased to say that openafs works like a champ. Sadly, as Chaskiel and Carson deduced I am not in a position to run an entirely 64bit user space. I wish I was. My goal is that running a 32bit user space, with a few statically linked binaries will provide a transition to a full 64 bit environment.. Thanks for your help, Ken ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
Re: [OpenAFS] openafs w/ amd64 kernel and 32bit user space
In message [EMAIL PROTECTED],Carson Gaspar writes: And sadly many user-land pieces still are not 64-bit clean. I know that I've had to fall back to 32-bit mode for several apps on my Solaris x86 server, as I didn't have the time energy to fix the crappy source. i know there are issues with running a 64-bit server on ia64. i suspect its mostly just missing prototypes breaking bosserver et al. fixing prototypes was the bulk of the work to get the afs client working on ia64. if you could mention which apps (or fill a bug report) someone might work on them. ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
[OpenAFS] openafs w/ amd64 kernel and 32bit user space
I am hoping to run a mixed environment with an amd64 kernel and a 32bit user space (along with a few statically compiled 64 bit apps). Before I try this I would like to know if the openafs user space tools running in the 32bit user space will communicate correctly with the amd64 kernel? Has anyone tried this? Thanks, Ken ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
Re: [OpenAFS] openafs w/ amd64 kernel and 32bit user space
Ken: On Fri, 23 Mar 2007, Kenneth Baker wrote: I am hoping to run a mixed environment with an amd64 kernel and a 32bit user space (along with a few statically compiled 64 bit apps). Before I try this I would like to know if the openafs user space tools running in the 32bit user space will communicate correctly with the amd64 kernel? Has anyone tried this? Yes, this works; however, why do you want to use 32-bit user space? The general consensus is that AMD64 code is faster/better in many ways and you really shouldn't use i386 code on a 64-bit CPU if you can avoid it. (this differs from e.g. Sparc where sparc32 code is generally faster than sparc64 code, and so it makes sense to use a 32-bit userland there) Thanks, Chris Wing [EMAIL PROTECTED] ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
Re: [OpenAFS] openafs w/ amd64 kernel and 32bit user space
--On Friday, March 23, 2007 12:33:42 PM -0400 Christopher Allen Wing [EMAIL PROTECTED] wrote: Yes, this works; however, why do you want to use 32-bit user space? Not that I know ken's reasoning, but here's why I would do so: Because I already have a managed 32-bit environment, and creating a 64-bit version of that environment would take more effort than I can afford to spend (but building and testing a kernel and some kernel modules would not) ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
Re: [OpenAFS] openafs w/ amd64 kernel and 32bit user space
Chaskiel M Grundman wrote: --On Friday, March 23, 2007 12:33:42 PM -0400 Christopher Allen Wing [EMAIL PROTECTED] wrote: Yes, this works; however, why do you want to use 32-bit user space? Not that I know ken's reasoning, but here's why I would do so: Because I already have a managed 32-bit environment, and creating a 64-bit version of that environment would take more effort than I can afford to spend (but building and testing a kernel and some kernel modules would not) And sadly many user-land pieces still are not 64-bit clean. I know that I've had to fall back to 32-bit mode for several apps on my Solaris x86 server, as I didn't have the time energy to fix the crappy source. -- Carson ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
