Re: yet another bug in atyfb.c

On Tue, Aug 03, 1999 at 11:26:39AM -0400, David Edelsohn wrote:

quoted

quoted

quoted

quoted

quoted

Wulf Hofbauer writes:

Wulf> EXPLANATION: atyfb.c - as of Kernel 2.2.10 - uses the following constructs
Wulf> for accessing little-endian words in Mach32 controller space:

Wulf> asm("lwbrx %0,%1,%2" : "=r"(val) : "r" (regindex), "r" (temp));

Wulf> and

Wulf> asm("stwbrx %0,%1,%2" : : "r" (val), "r" (regindex), "r" (temp) :
Wulf> "memory");

Wulf> This is meant to access a word at address regindex+temp. If regindex
Wulf> happens to be held in register r0, the address calculation is off as
Wulf> r0 is defined as a null operand. This problem shows up with gcc-2.95 which
Wulf> seems to use better register allocation code and _does_ keep regindex in
Wulf> r0 at times.

	The problem is that you are using the wrong register constraints.
The inlined assembly should look like:

asm("lwbrx %0,%1,%2" : "=r"(val) : "b" (regindex), "r" (temp));

because %1 must be a BASE register (any GPR other than r0) for this use.
If you use the correct register constraints, GCC register allocation will
arrange to place the values in the correct class of register.  If you use
the right register constraints, you do not need to make any other
modifications. 

And congratulations to the both of you, you analyzed that one much more
easily than I did :)  It's been fixed in vger for a while:

revision 1.106.2.1
date: 1999/06/22 06:28:23;  author: paulus;  state: Exp;  lines: +42 -23
Geert's crystal frequency estimation stuff from the 2.3 branch;
fix the constraints on the inline assembly.

However, could someone please move this fix onto the mainline also?


Dan

/--------------------------------\  /--------------------------------\
|       Daniel Jacobowitz        |__|        SCS Class of 2002       |
|   Debian GNU/Linux Developer    __    Carnegie Mellon University   |
|         dan@debian.org         |  |       dmj+@andrew.cmu.edu      |
\--------------------------------/  \--------------------------------/

[[ This message was sent via the linuxppc-dev mailing list.  Replies are ]]
[[ not  forced  back  to the list, so be sure to Cc linuxppc-dev if your ]]
[[ reply is of general interest. Please check http://lists.linuxppc.org/ ]]
[[ and http://www.linuxppc.org/ for useful information before posting.   ]]