Maximum Alignment Restrictions

Maximum Alignment Restrictions

[Joel Sherrill]

What is the rationale for the maximum alignment supported on a particular 
CPU family?  

My current problem is that the SPARC trap table must be on a 4K boundary
but that appears to be too large. 

My thinking is that there is no particular reason (except object format 
limitations) to place a CPU dependent limit on maximum alignment boundary.

Just curious.

+----------------------------------------+--------------------------------+
| Joel Sherrill                          | Sr. Computer Scientist         |
| joel@merlin.gcs.redstone.army.mil      | On-Line Applications Research  |
| Ask me about RTEMS: a free real-time   | Huntsville AL 35805            |
|   multiprocessor executive!            | (205) 883-0131                 |
+----------------------------------------+--------------------------------+

Re: Maximum Alignment Restrictions

[Ian Lance Taylor]

   Date: Thu, 11 Jan 1996 11:32:55 -0600 (CST)
   From: Joel Sherrill <joel@merlin.gcs.redstone.army.mil>

   What is the rationale for the maximum alignment supported on a particular 
   CPU family?  

There is no such maximum in BFD.  In BFD, a CPU type provides a
default alignment.

   My current problem is that the SPARC trap table must be on a 4K boundary
   but that appears to be too large. 

   My thinking is that there is no particular reason (except object format 
   limitations) to place a CPU dependent limit on maximum alignment boundary.

There is no other reason.

In what way is this affecting you?  You didn't say what target you are
using.  Perhaps you are using a SPARC a.out target.  The problem there
is the a.out object file format has no way to specify the alignment of
a section.  BFD must guess, and it guesses based on the CPU type.  The
CPU type is not specifying a maximum alignment.  It is providing a
default alignment to use when the alignment can not be otherwise
determined.

Ian

Re: Maximum Alignment Restrictions

[David S. Miller]

   Date: Thu, 11 Jan 1996 11:32:55 -0600 (CST)
   From: Joel Sherrill <joel@merlin.gcs.redstone.army.mil>

   What is the rationale for the maximum alignment supported on a particular 
   CPU family?  

   My current problem is that the SPARC trap table must be on a 4K boundary
   but that appears to be too large. 

Note that many of these restrictions eminate from the
linkers/assemblers which gcc has to interface with.

As for your trap table alignment problem, I have no problem getting
mine properly aligned in my kernels at all.  I know where I am going
to fix the text start address in the final link, so I know where the
text segment will be in the first object file I feed to the linker.
So this is where I place the trap table etc... this is what all Sparc
OS's do btw, check out netbsd/openbsd or Sparc Linux for example.  It
is not really a problem.

Later,
David S. Miller
davem@caip.rutgers.edu