>>>>> "Peter" == Peter Danielsson <[EMAIL PROTECTED]> writes:
Peter> Peter Teichmann wrote:
>> Well I compiled the kernel and the modules by myself, but I did
>> not know that I need a new FPEmulator. So I downloaded it from
>> ftp.arm.linux.org.uk. Now it complains no more about wrong
>> versions, but does not print something about the fpe module loaded
>> as the old kernel did. It also does not let me login. I also
>> included nwfpe as module, but it did not help. I also compiled a
>> kernel with all things included, not as modules, but that had also
>> no effect. If I tell you that compiling a kernel lasts about 1.5
>> hours on my RiscPC you know that I wasted a lot of time for
>> nothing.
>>
Peter> I recognize this problem! I too had some troubles with login,
Peter> and it seemed impossible to get a working version (I compiled
Peter> two or three myself, downloaded precompiled binaries and
Peter> nothing helped.) After a while I begun to supect glibc for
Peter> having a bug so I rewrote login to use some other routines for
Peter> accessing /etc/passwd. ...
That rings a bell (it has been ringing in the background for a while,
actually).
There was a discussion on this list only a week or two ago about a bug
in the glibc password handling functions -- caused by an invalid
assumption about alignment of data elements in structures. Solution
was to mark the structure in question "packed". A scan of the
archives should turn it up soon enough...
I wonder how many other cases like this exist. It's scary to think of
the number of source lines that might hide bugs of this kind. Perhaps
a solution would be to add a warning feature to the compiler so it
reports any structure declarations where the particular platform is
using padding beyond the minimum. I don't speak enough gcc to offer
to do it, unfortunately.
paul
unsubscribe: body of `unsubscribe linux-arm' to [EMAIL PROTECTED]
