Use the ABI-mandated thread pointer register (r2 for ppc32, r13 for ppc64)

instead of a PCPU field for curthread. This averts a race on SMP systems
with a high interrupt rate where the thread looking up the value of
curthread could be preempted and migrated between obtaining the PCPU
pointer and reading the value of pc_curthread, resulting in curthread being
observed to be the current thread on the thread's original CPU. This played
merry havoc with the system, in particular with mutexes. Many thanks to
jhb for helping me work this one out.

Note that Book-E is in principle susceptible to the same problem, but has
not been modified yet due to lack of Book-E hardware.

MFC after:	2 weeks

1 files changed, 8 insertions, 10 deletions

diff --git a/sys/powerpc/aim/swtch64.S b/sys/powerpc/aim/swtch64.S
index 1877e8d..c5cdcbc 100644
--- a/sys/powerpc/aim/swtch64.S
+++ b/sys/powerpc/aim/swtch64.S
@@ -69,7 +69,7 @@
  * void cpu_throw(struct thread *old, struct thread *new)
  */
 ENTRY(cpu_throw)
-	mr	%r15, %r4
+	mr	%r13, %r4
 	b	cpu_switchin
 
 /*
@@ -83,7 +83,6 @@ ENTRY(cpu_switch)
 	ld	%r6,TD_PCB(%r3)		/* Get the old thread's PCB ptr */
 	std	%r12,PCB_CONTEXT(%r6)	/* Save the non-volatile GP regs.
 					   These can now be used for scratch */
-	std	%r13,PCB_CONTEXT+1*8(%r6)	
 	std	%r14,PCB_CONTEXT+2*8(%r6)	
 	std	%r15,PCB_CONTEXT+3*8(%r6)	
 	std	%r16,PCB_CONTEXT+4*8(%r6)	
@@ -111,7 +110,7 @@ ENTRY(cpu_switch)
 	std	%r2,PCB_TOC(%r6)	/* Save the TOC pointer */
 	
 	mr	%r14,%r3		/* Copy the old thread ptr... */
-	mr	%r15,%r4		/* and the new thread ptr in scratch */
+	mr	%r13,%r4		/* and the new thread ptr in curthread*/
 	mr	%r16,%r5		/* and the new lock */
 	mr	%r17,%r6		/* and the PCB */
 	
@@ -148,19 +147,19 @@ cpu_switchin:
 	lis	%r6,blocked_lock@ha
 	addi	%r6,%r6,blocked_lock@l
 blocked_loop:
-	ld	%r7,TD_LOCK(%r15)
+	ld	%r7,TD_LOCK(%r13)
 	cmpd	%r6,%r7 
 	beq	blocked_loop
 #endif
 
 	mfsprg	%r7,0			/* Get the pcpu pointer */
-	std	%r15,PC_CURTHREAD(%r7)	/* Store new current thread */
-	ld	%r17,TD_PCB(%r15)	/* Store new current PCB */
+	std	%r13,PC_CURTHREAD(%r7)	/* Store new current thread */
+	ld	%r17,TD_PCB(%r13)	/* Store new current PCB */
 	std	%r17,PC_CURPCB(%r7)
 
 	stdu	%r1,-48(%r1)
 
-	mr	%r3,%r15		/* Get new thread ptr */
+	mr	%r3,%r13		/* Get new thread ptr */
 	bl	pmap_activate		/* Activate the new address space */
 	nop
 
@@ -168,7 +167,7 @@ blocked_loop:
 	/* Restore FPU context if needed */
 	andi.	%r6, %r6, PCB_FPU
 	beq	.L3
-	mr	%r3,%r15		/* Pass curthread to enable_fpu */
+	mr	%r3,%r13		/* Pass curthread to enable_fpu */
 	bl	enable_fpu
 	nop
 
@@ -177,7 +176,7 @@ blocked_loop:
 	/* Restore Altivec context if needed */
 	andi.	%r6, %r6, PCB_VEC
 	beq	.L4
-	mr	%r3,%r15		/* Pass curthread to enable_vec */
+	mr	%r3,%r13		/* Pass curthread to enable_vec */
 	bl	enable_vec
 	nop
 
@@ -186,7 +185,6 @@ blocked_loop:
 	addi	%r1,%r1,48
 	mr	%r3,%r17		/* Recover PCB ptr */
 	ld	%r12,PCB_CONTEXT(%r3)	/* Load the non-volatile GP regs. */
-	ld	%r13,PCB_CONTEXT+1*8(%r3)	
 	ld	%r14,PCB_CONTEXT+2*8(%r3)	
 	ld	%r15,PCB_CONTEXT+3*8(%r3)	
 	ld	%r16,PCB_CONTEXT+4*8(%r3)