[Bug runtime/12341] New: deref()/store_deref() variations between arches

[Bug runtime/12341] New: deref()/store_deref() variations between arches

[dsmith at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=12341

           Summary: deref()/store_deref() variations between arches
           Product: systemtap
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: runtime
        AssignedTo: systemtap@sources.redhat.com
        ReportedBy: dsmith@redhat.com


There is a different deref()/store_deref() macro in runtime/loc2c-runtime.h for
each arch.

For i386/x86_64, deref() calls the kernel's __get_user_asm().  store_deref()
calls the kernel's __put_user_asm().

For ia64, deref() calls the kernel's __get_user_size().  store_deref() calls
the kernel's __put_user_size().

For powerpc, deref() calls __stp_get_user_asm(), which is custom powerpc
assembly code (that looks almost identical to the kernel's __get_user_asm()). 
store_deref() calls __stp_put_user_asm(), which is custom powerpc assembly
code.

For s390, deref() calls __stp_get_user_asm(), which is custom s390 assembly. 
store_deref() calls __stp_put_user_asm(), which is custom s390 assembly.

It might be possible for all arches to use the kernel's own routines, instead
of having custom (powerpc/s390) assembly code.  The
__get_user_size()/__put_user_size() routines might be the most attractive
choice (as the ia64 uses), since they know how handle all the different size
combinations.

(If __get_user_size()/__put_user_size() works for the i386, that could
eliminate the need for the i386 special version of kread/kwrite that can handle
64-bit values.)

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug runtime/12341] deref()/store_deref() variations between arches

[dsmith at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=12341

--- Comment #1 from David Smith <dsmith at redhat dot com> 2011-01-04 21:50:16 UTC ---
(In reply to comment #0)
> (If __get_user_size()/__put_user_size() works for the i386, that could
> eliminate the need for the i386 special version of kread/kwrite that can handle
> 64-bit values.)

Unfortunately, this idea won't work.  Looking at the i386 header files in
kernels 2.6.9/2.6.18/2.6.36, none of the i386 __get_user_size() macros can
handle 64-bit values.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug runtime/12341] deref()/store_deref() variations between arches

[dsmith at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=12341

--- Comment #2 from David Smith <dsmith at redhat dot com> 2011-01-06 20:58:39 UTC ---
(In reply to comment #0)
> For powerpc, deref() calls __stp_get_user_asm(), which is custom powerpc
> assembly code (that looks almost identical to the kernel's __get_user_asm()). 
> store_deref() calls __stp_put_user_asm(), which is custom powerpc assembly
> code.

For powerpc, I've discovered some history.

loc2c-runtime.h originally used the kernel's __get_user_asm() macros.

This changed on 20 Mar 2006 in the following commit:

<http://sources.redhat.com/git/gitweb.cgi?p=systemtap.git;a=commitdiff;h=15eeafe9465119eb443f59b332e24ecfeec10488>

To fix bug 2076 (ppc64 runtime is broken under kernel v2.6.15-RC5):

<http://sourceware.org/bugzilla/show_bug.cgi?id=2076>

which was caused by the "__get_user_asm" macro changing to taking 4 arguments
(from 5).

There is (hopefully) a better way to fix that problem than duplicating
__get_user_asm()/__put_user_asm().

Note that we just can't go back to calling __get_user_asm(), since in current
kernels __get_user_asm2() is used on powerpc to get 64-bit values.  We'll need
to call something a bit higher up.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.