[PATCH] spufs: Fix endless protection fault on LS writes by SPE.

If an SPE attempts a DMA put to a local store after already doing
a get, the kernel must update the HW PTE to allow the write access.
This case was not being handled correctly.

From: Mike Kistler <mkistler@us.ibm.com>
Signed-off-by: NMike Kistler <mkistler@us.ibm.com>
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

arch/powerpc/platforms/cell/spu_base.c

@@ -486,14 +486,13 @@ int spu_irq_class_1_bottom(struct spu *spu)
 
 	ea = spu->dar;
 	dsisr = spu->dsisr;
-	if (dsisr & MFC_DSISR_PTE_NOT_FOUND) {
+	if (dsisr & (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED)) {
 		access = (_PAGE_PRESENT | _PAGE_USER);
 		access |= (dsisr & MFC_DSISR_ACCESS_PUT) ? _PAGE_RW : 0UL;
 		if (hash_page(ea, access, 0x300) != 0)
 			error |= CLASS1_ENABLE_STORAGE_FAULT_INTR;
 	}
-	if ((error & CLASS1_ENABLE_STORAGE_FAULT_INTR) ||
-	    (dsisr & MFC_DSISR_ACCESS_DENIED)) {
+	if (error & CLASS1_ENABLE_STORAGE_FAULT_INTR) {
 		if ((ret = spu_handle_mm_fault(spu)) != 0)
 			error |= CLASS1_ENABLE_STORAGE_FAULT_INTR;
 		else