floating point register corruption

[Paolo Galtieri]

Folks,
  I recently discovered a bug on PPC which causes the floating
point registers to get corrupted when CONFIG_PREEMPT=y.

The problem occurred while running a multi threaded Java application
that does floating point.  The problem could be reproduced in
anywhere from 2 to 6 hours.  With the patch I have included below
it ran for over a week without failure.

Paolo

Signed-off-by: pgaltieri@xxxxxxxxxx

--- linux-2.6.15-rc3/arch/ppc/kernel/process.c	2005-11-29
07:01:55.000000000 -0700
+++ new-linux-2.6.15-rc3/arch/ppc/kernel/process.c	2005-11-29
07:20:37.000000000 -0700
@@ -417,6 +417,7 @@
 
 void exit_thread(void)
 {
+	preempt_disable();
 	if (last_task_used_math == current)
 		last_task_used_math = NULL;
 	if (last_task_used_altivec == current)
@@ -425,10 +426,12 @@
 	if (last_task_used_spe == current)
 		last_task_used_spe = NULL;
 #endif
+	preempt_enable();
 }
 
 void flush_thread(void)
 {
+	preempt_disable();
 	if (last_task_used_math == current)
 		last_task_used_math = NULL;
 	if (last_task_used_altivec == current)
@@ -437,6 +440,7 @@
 	if (last_task_used_spe == current)
 		last_task_used_spe = NULL;
 #endif
+	preempt_enable();
 }
 
 void
@@ -535,6 +539,7 @@
 	regs->nip = nip;
 	regs->gpr[1] = sp;
 	regs->msr = MSR_USER;
+	preempt_disable();
 	if (last_task_used_math == current)
 		last_task_used_math = NULL;
 	if (last_task_used_altivec == current)
@@ -543,6 +548,7 @@
 	if (last_task_used_spe == current)
 		last_task_used_spe = NULL;
 #endif
+	preempt_enable();
 	memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 	current->thread.fpscr.val = 0;
 #ifdef CONFIG_ALTIVEC


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/