Experimental Physics and
Industrial Control System

"Lawrence T. Hoff" <[email protected]> · Wed, 29 Sep 2004 05:57:53 -0400

	Among other things, -mstrict-align prevents the compiler 
from using a (64-bit) floating point register to store a 64-bit
integer structure. If the compiler uses a FP register for 
integer data, it can result in unpredictable results when
context switching to/from tasks which are not designated as FP
tasks in VxWorks. If your O/S does not distinguish between FP
and non-FP tasks, then this issue is moot. An example of a 64-bit
integer data structure is the time data structure used in the
function "select". 

	I.e. this code is vulnerable if compiled without -mstrict-align,
and used in VxWorks as a non-FP task:

for(;;){

     struct timeval oneSecond = { 1, 0 };

     select(1, NULL, NULL, NULL, &oneSecond);

     doOneSecondTask();

     }

	If the compiler chooses to store "oneSecond" in a FP register
for convenience, the contents can get "clobbered" by a true FP task
using that register for FP purposes, because VxWorks will not restore
FP registers when switching back to this non-FP task.

	Bottom line : I recommend continuing to compile with
-mstrict-align
or configuring *all* "spawned" tasks as FP.

HTH -- Larry

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System