x86: Handle HW IOMMU initialization failure gracefully

If HW IOMMU initialization fails (Intel VT-d often does this,
typically due to BIOS bugs), we fall back to nommu. It doesn't
work for the majority since nowadays we have more than 4GB
memory so we must use swiotlb instead of nommu.

The problem is that it's too late to initialize swiotlb when HW
IOMMU initialization fails. We need to allocate swiotlb memory
earlier from bootmem allocator. Chris explained the issue in
detail:

  http://marc.info/?l=linux-kernel&m=125657444317079&w=2

The current x86 IOMMU initialization sequence is too complicated
and handling the above issue makes it more hacky.

This patch changes x86 IOMMU initialization sequence to handle
the above issue cleanly.

The new x86 IOMMU initialization sequence are:

1. we initialize the swiotlb (and setting swiotlb to 1) in the case
   of (max_pfn > MAX_DMA32_PFN && !no_iommu). dma_ops is set to
   swiotlb_dma_ops or nommu_dma_ops. if swiotlb usage is forced by
   the boot option, we finish here.

2. we call the detection functions of all the IOMMUs

3. the detection function sets x86_init.iommu.iommu_init to the
   IOMMU initialization function (so we can avoid calling the
   initialization functions of all the IOMMUs needlessly).

4. if the IOMMU initialization function doesn't need to swiotlb
   then sets swiotlb to zero (e.g. the initialization is
   sucessful).

5. if we find that swiotlb is set to zero, we free swiotlb
   resource.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: chrisw@sous-sol.org
Cc: dwmw2@infradead.org
Cc: joerg.roedel@amd.com
Cc: muli@il.ibm.com
LKML-Reference: <1257849980-22640-10-git-send-email-fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

9 files changed, 21 insertions, 34 deletions

diff --git a/arch/x86/include/asm/iommu.h b/arch/x86/include/asm/iommu.h
index 878b30715766..df42a712361f 100644
--- a/arch/x86/include/asm/iommu.h
+++ b/arch/x86/include/asm/iommu.h
@@ -2,7 +2,6 @@
 #define _ASM_X86_IOMMU_H
 
 static inline void iommu_shutdown_noop(void) {}
-extern void no_iommu_init(void);
 extern struct dma_map_ops nommu_dma_ops;
 extern int force_iommu, no_iommu;
 extern int iommu_detected;
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 0285521e0a99..66237fde758f 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -2110,8 +2110,8 @@ int __init amd_iommu_init_dma_ops(void)
                 prealloc_protection_domains();
 
         iommu_detected = 1;
-        force_iommu = 1;
         bad_dma_address = 0;
+        swiotlb = 0;
 #ifdef CONFIG_GART_IOMMU
         gart_iommu_aperture_disabled = 1;
         gart_iommu_aperture = 0;
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index c41aabddaa2a..0d4581e602a4 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -1330,7 +1330,7 @@ static int __init early_amd_iommu_detect(struct acpi_table_header *table)
 
 void __init amd_iommu_detect(void)
 {
-        if (swiotlb || no_iommu || (iommu_detected && !gart_iommu_aperture))
+        if (no_iommu || (iommu_detected && !gart_iommu_aperture))
                 return;
 
         if (acpi_table_parse("IVRS", early_amd_iommu_detect) == 0) {
diff --git a/arch/x86/kernel/aperture_64.c b/arch/x86/kernel/aperture_64.c
index 03933cf0b63c..e0dfb6856aa2 100644
--- a/arch/x86/kernel/aperture_64.c
+++ b/arch/x86/kernel/aperture_64.c
@@ -458,7 +458,7 @@ out:
 
         if (aper_alloc) {
                 /* Got the aperture from the AGP bridge */
-        } else if (swiotlb && !valid_agp) {
+        } else if (!valid_agp) {
                 /* Do nothing */
         } else if ((!no_iommu && max_pfn > MAX_DMA32_PFN) ||
                    force_iommu ||
diff --git a/arch/x86/kernel/pci-calgary_64.c b/arch/x86/kernel/pci-calgary_64.c
index 47bd419ea4d2..833f491440b9 100644
--- a/arch/x86/kernel/pci-calgary_64.c
+++ b/arch/x86/kernel/pci-calgary_64.c
@@ -1360,7 +1360,7 @@ void __init detect_calgary(void)
          * if the user specified iommu=off or iommu=soft or we found
          * another HW IOMMU already, bail out.
          */
-        if (swiotlb || no_iommu || iommu_detected)
+        if (no_iommu || iommu_detected)
                 return;
 
         if (!use_calgary)
@@ -1445,10 +1445,6 @@ void __init detect_calgary(void)
                 printk(KERN_INFO "PCI-DMA: Calgary TCE table spec is %d\n",
                        specified_table_size);
 
-                /* swiotlb for devices that aren't behind the Calgary. */
-                if (max_pfn > MAX_DMA32_PFN)
-                        swiotlb = 1;
-
                 x86_init.iommu.iommu_init = calgary_iommu_init;
         }
         return;
@@ -1476,11 +1472,7 @@ int __init calgary_iommu_init(void)
                 return ret;
         }
 
-        force_iommu = 1;
         bad_dma_address = 0x0;
-        /* dma_ops is set to swiotlb or nommu */
-        if (!dma_ops)
-                dma_ops = &nommu_dma_ops;
 
         return 0;
 }
diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c
index bed05e2e5890..a234e63c2656 100644
--- a/arch/x86/kernel/pci-dma.c
+++ b/arch/x86/kernel/pci-dma.c
@@ -124,24 +124,24 @@ static void __init dma32_free_bootmem(void)
 
 void __init pci_iommu_alloc(void)
 {
+        /* swiotlb is forced by the boot option */
+        int use_swiotlb = swiotlb;
 #ifdef CONFIG_X86_64
         /* free the range so iommu could get some range less than 4G */
         dma32_free_bootmem();
 #endif
+        pci_swiotlb_init();
+        if (use_swiotlb)
+                return;
 
-        /*
-         * The order of these functions is important for
-         * fall-back/fail-over reasons
-         */
         gart_iommu_hole_init();
 
         detect_calgary();
 
         detect_intel_iommu();
 
+        /* needs to be called after gart_iommu_hole_init */
         amd_iommu_detect();
-
-        pci_swiotlb_init();
 }
 
 void *dma_generic_alloc_coherent(struct device *dev, size_t size,
@@ -291,10 +291,15 @@ static int __init pci_iommu_init(void)
 #ifdef CONFIG_PCI
         dma_debug_add_bus(&pci_bus_type);
 #endif
-
         x86_init.iommu.iommu_init();
 
-        no_iommu_init();
+        if (swiotlb) {
+                printk(KERN_INFO "PCI-DMA: "
+                       "Using software bounce buffering for IO (SWIOTLB)\n");
+                swiotlb_print_info();
+        } else
+                swiotlb_free();
+
         return 0;
 }
 /* Must execute after PCI subsystem */
diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c
index 0410bd30060d..919182e15d1e 100644
--- a/arch/x86/kernel/pci-gart_64.c
+++ b/arch/x86/kernel/pci-gart_64.c
@@ -833,6 +833,7 @@ int __init gart_iommu_init(void)
         flush_gart();
         dma_ops = &gart_dma_ops;
         x86_platform.iommu_shutdown = gart_iommu_shutdown;
+        swiotlb = 0;
 
         return 0;
 }
diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c
index a3933d4330cd..875e3822ae61 100644
--- a/arch/x86/kernel/pci-nommu.c
+++ b/arch/x86/kernel/pci-nommu.c
@@ -103,12 +103,3 @@ struct dma_map_ops nommu_dma_ops = {
         .sync_sg_for_device     = nommu_sync_sg_for_device,
         .is_phys                = 1,
 };
-
-void __init no_iommu_init(void)
-{
-        if (dma_ops)
-                return;
-
-        force_iommu = 0; /* no HW IOMMU */
-        dma_ops = &nommu_dma_ops;
-}
diff --git a/arch/x86/kernel/pci-swiotlb.c b/arch/x86/kernel/pci-swiotlb.c
index ea20ef7ca523..17ce4221bd03 100644
--- a/arch/x86/kernel/pci-swiotlb.c
+++ b/arch/x86/kernel/pci-swiotlb.c
@@ -46,13 +46,12 @@ void __init pci_swiotlb_init(void)
 {
         /* don't initialize swiotlb if iommu=off (no_iommu=1) */
 #ifdef CONFIG_X86_64
-        if ((!iommu_detected && !no_iommu && max_pfn > MAX_DMA32_PFN))
+        if (!no_iommu && max_pfn > MAX_DMA32_PFN)
                 swiotlb = 1;
 #endif
-        if (swiotlb_force)
-                swiotlb = 1;
         if (swiotlb) {
                 swiotlb_init(0);
                 dma_ops = &swiotlb_dma_ops;
-        }
+        } else
+                dma_ops = &nommu_dma_ops;
 }