x86, AMD IOMMU: use status bit instead of memory write-back for completion wait

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 22d7d05..028e945 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -101,16 +101,13 @@
  */
 static int iommu_completion_wait(struct amd_iommu *iommu)
 {
-	int ret;
+	int ret, ready = 0;
+	unsigned status = 0;
 	struct iommu_cmd cmd;
-	volatile u64 ready = 0;
-	unsigned long ready_phys = virt_to_phys(&ready);
 	unsigned long i = 0;
 
 	memset(&cmd, 0, sizeof(cmd));
-	cmd.data[0] = LOW_U32(ready_phys) | CMD_COMPL_WAIT_STORE_MASK;
-	cmd.data[1] = upper_32_bits(ready_phys);
-	cmd.data[2] = 1; /* value written to 'ready' */
+	cmd.data[0] = CMD_COMPL_WAIT_INT_MASK;
 	CMD_SET_TYPE(&cmd, CMD_COMPL_WAIT);
 
 	iommu->need_sync = 0;
@@ -122,9 +119,15 @@
 
 	while (!ready && (i < EXIT_LOOP_COUNT)) {
 		++i;
-		cpu_relax();
+		/* wait for the bit to become one */
+		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
+		ready = status & MMIO_STATUS_COM_WAIT_INT_MASK;
 	}
 
+	/* set bit back to zero */
+	status &= ~MMIO_STATUS_COM_WAIT_INT_MASK;
+	writel(status, iommu->mmio_base + MMIO_STATUS_OFFSET);
+
 	if (unlikely((i == EXIT_LOOP_COUNT) && printk_ratelimit()))
 		printk(KERN_WARNING "AMD IOMMU: Completion wait loop failed\n");