[BUG & PATCH V2] drivers/pci/intel-iommu.c: errors with smalleriommu widths

[Tom Lyon]

When using iommu_domain_alloc with the Intel iommu, the domain address width
is always initialized to 48 bits (agaw 2). ÂThis domain->agaw value is then
used by pfn_to_dma_pte to (always) build a 4 level page table. ÂHowever, not
all systems support iommu width of 48 or 4 level page tables. ÂIn particular,
the Core i5-660 and i5-670 support an address width of 36 bits (not 39!), an
agaw of only 1, and only 3 level page tables.

This version of the patch simply lops off extra levels of the page tables if
the agaw value of the iommu is less than what is currently allocated for the
domain (in intel_iommu_attach_device). If there were already allocated
addresses above what the new iommu can handle, EFAULT is returned.

A related bug in intel_iommu_map_range didn't allow allocation at the very
end of the address space, that code has been simplified and corrected.

--- linux-2.6.33/drivers/pci/intel-iommu.c	2010-02-24 10:52:17.000000000 -0800
+++ mylinux-2.6.33/drivers/pci/intel-iommu.c	2010-04-13 16:51:55.000000000 -0700
@@ -3436,22 +3436,6 @@
 /* domain id for virtual machine, it won't be set in context */
 static unsigned long vm_domid;
 
-static int vm_domain_min_agaw(struct dmar_domain *domain)
-{
-	int i;
-	int min_agaw = domain->agaw;
-
-	i = find_first_bit(&domain->iommu_bmp, g_num_of_iommus);
-	for (; i < g_num_of_iommus; ) {
-		if (min_agaw > g_iommus[i]->agaw)
-			min_agaw = g_iommus[i]->agaw;
-
-		i = find_next_bit(&domain->iommu_bmp, g_num_of_iommus, i+1);
-	}
-
-	return min_agaw;
-}
-
 static struct dmar_domain *iommu_alloc_vm_domain(void)
 {
 	struct dmar_domain *domain;
@@ -3582,7 +3566,6 @@
 	struct pci_dev *pdev = to_pci_dev(dev);
 	struct intel_iommu *iommu;
 	int addr_width;
-	u64 end;
 
 	/* normally pdev is not mapped */
 	if (unlikely(domain_context_mapped(pdev))) {
@@ -3605,14 +3588,30 @@
 
 	/* check if this iommu agaw is sufficient for max mapped address */
 	addr_width = agaw_to_width(iommu->agaw);
-	end = DOMAIN_MAX_ADDR(addr_width);
-	end = end & VTD_PAGE_MASK;
-	if (end < dmar_domain->max_addr) {
-		printk(KERN_ERR "%s: iommu agaw (%d) is not "
+	if (addr_width > cap_mgaw(iommu->cap))
+		addr_width = cap_mgaw(iommu->cap);
+
+	if (dmar_domain->max_addr > (1LL << addr_width)) {
+		printk(KERN_ERR "%s: iommu width (%d) is not "
 		       "sufficient for the mapped address (%llx)\n",
-		       __func__, iommu->agaw, dmar_domain->max_addr);
+		       __func__, addr_width, dmar_domain->max_addr);
 		return -EFAULT;
 	}
+	dmar_domain->gaw = addr_width;
+
+	/*
+	 * Knock out extra levels of page tables if necessary
+	 */
+	while (iommu->agaw < dmar_domain->agaw) {
+		struct dma_pte *pte;
+
+		pte = dmar_domain->pgd;
+		if (dma_pte_present(pte)) {
+			free_pgtable_page(dmar_domain->pgd);
+			dmar_domain->pgd = (struct dma_pte *)dma_pte_addr(pte);
+		}
+		dmar_domain->agaw--;
+	}
 
 	return domain_add_dev_info(dmar_domain, pdev, CONTEXT_TT_MULTI_LEVEL);
 }
@@ -3632,7 +3631,6 @@
 {
 	struct dmar_domain *dmar_domain = domain->priv;
 	u64 max_addr;
-	int addr_width;
 	int prot = 0;
 	int ret;
 
@@ -3645,18 +3643,14 @@
 
 	max_addr = iova + size;
 	if (dmar_domain->max_addr < max_addr) {
-		int min_agaw;
 		u64 end;
 
 		/* check if minimum agaw is sufficient for mapped address */
-		min_agaw = vm_domain_min_agaw(dmar_domain);
-		addr_width = agaw_to_width(min_agaw);
-		end = DOMAIN_MAX_ADDR(addr_width);
-		end = end & VTD_PAGE_MASK;
+		end = __DOMAIN_MAX_ADDR(dmar_domain->gaw) + 1;
 		if (end < max_addr) {
-			printk(KERN_ERR "%s: iommu agaw (%d) is not "
+			printk(KERN_ERR "%s: iommu width (%d) is not "
 			       "sufficient for the mapped address (%llx)\n",
-			       __func__, min_agaw, max_addr);
+			       __func__, dmar_domain->gaw, max_addr);
 			return -EFAULT;
 		}
 		dmar_domain->max_addr = max_addr;
--
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/