Skip to contentSkip to conversation

amd iommu fixes for 2.6.35-rc1

GIT PULL New Joerg Roedel 2 participants 5 messages Jun 1, 2010 x86

lore
Share & tools
lore
#0 lore raw

Hi Ingo,

The following changes since commit 67a3e12b05e055c0415c556a315a3d3eb637e29e:

Linux 2.6.35-rc1 (2010-05-30 13:21:02 -0700)

are available in the git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/joro/linux-2.6-iommu.git amd-iommu/2.6.35

Joerg Roedel (3):
Merge commit 'v2.6.35-rc1' into amd-iommu/2.6.35 x86/amd-iommu: Fix crash when request_mem_region fails x86/amd-iommu: Fall back to GART if initialization fails

Julia Lawall (1):
arch/x86/kernel: Add missing spin_unlock

 arch/x86/kernel/amd_iommu.c      |   16 +++++++++-------
 arch/x86/kernel/amd_iommu_init.c |   20 +++++++++++++++++---
 2 files changed, 26 insertions(+), 10 deletions(-)

These patches fix two bugs in the error-paths of the amd iommu driver. One fix adds missing spin_unlocks and the other two patches fix a problem with broken BIOSes where the driver can't allocate the mmio region for the iommu hardware. Please pull.

Joerg

#1 lore raw

PATCH 1/3 arch/x86/kernel: Add missing spin_unlock

From: Julia Lawall <[email protected]>

Add a spin_unlock missing on the error path. The locks and unlocks are balanced in other functions, so it seems that the same should be the case here.

The semantic match that finds this problem is as follows:
(http://coccinelle.lip6.fr/)

// <smpl>
@@
expression E1;
@@

* spin_lock(E1,...);
<+... when != E1

  if (...) {
    ... when != E1

* return ...;
}
...+>
* spin_unlock(E1,...);
// </smpl>

---
 arch/x86/kernel/amd_iommu.c |   12 +++++++++---
 1 files changed, 9 insertions(+), 3 deletions(-)
arch/x86/kernel/amd_iommu.c +9 -3
index fa5a147..8a9aaa8 100644
@@ -1487,6 +1487,7 @@ static int __attach_device(struct device *dev,
 			   struct protection_domain *domain)
 {
 	struct iommu_dev_data *dev_data, *alias_data;
+	int ret;
 
 	dev_data   = get_dev_data(dev);
 	alias_data = get_dev_data(dev_data->alias);
@@ -1498,13 +1499,14 @@ static int __attach_device(struct device *dev,
 	spin_lock(&domain->lock);
 
 	/* Some sanity checks */
+	ret = -EBUSY;
 	if (alias_data->domain != NULL &&
 	    alias_data->domain != domain)
-		return -EBUSY;
+		goto out_unlock;
 
 	if (dev_data->domain != NULL &&
 	    dev_data->domain != domain)
-		return -EBUSY;
+		goto out_unlock;
 
 	/* Do real assignment */
 	if (dev_data->alias != dev) {
@@ -1520,10 +1522,14 @@ static int __attach_device(struct device *dev,
 
 	atomic_inc(&dev_data->bind);
 
+	ret = 0;
+
+out_unlock:
+
 	/* ready */
 	spin_unlock(&domain->lock);
 
-	return 0;
+	return ret;
 }
 
 /*
Signature · 1 line
1.7.1
#2 lore raw

PATCH 2/3 x86/amd-iommu: Fix crash when request_mem_region fails

In reply to Joerg Roedel

When request_mem_region fails the error path tries to disable the IOMMUs. This accesses the mmio-region which was not allocated leading to a kernel crash. This patch fixes the issue.

---
 arch/x86/kernel/amd_iommu_init.c |   11 ++++++++---
 1 files changed, 8 insertions(+), 3 deletions(-)
arch/x86/kernel/amd_iommu_init.c +8 -3
index 3bacb4d..1405346 100644
@@ -287,8 +287,12 @@ static u8 * __init iommu_map_mmio_space(u64 address)
 {
 	u8 *ret;
 
-	if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu"))
+	if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu")) {
+		pr_err("AMD-Vi: Can not reserve memory region %llx for mmio\n",
+			address);
+		pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
 		return NULL;
+	}
 
 	ret = ioremap_nocache(address, MMIO_REGION_LENGTH);
 	if (ret != NULL)
@@ -1314,7 +1318,7 @@ static int __init amd_iommu_init(void)
 		ret = amd_iommu_init_dma_ops();
 
 	if (ret)
-		goto free;
+		goto free_disable;
 
 	amd_iommu_init_api();
 
@@ -1332,9 +1336,10 @@ static int __init amd_iommu_init(void)
 out:
 	return ret;
 
-free:
+free_disable:
 	disable_iommus();
 
+free:
 	amd_iommu_uninit_devices();
 
 	free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
Signature · 1 line
1.7.1
#3 lore raw

PATCH 3/3 x86/amd-iommu: Fall back to GART if initialization fails

In reply to Joerg Roedel

This patch implements a fallback to the GART IOMMU if this is possible and the AMD IOMMU initialization failed. Otherwise the fallback would be nommu which is very problematic on machines with more than 4GB of memory or swiotlb which hurts io-performance.

---
 arch/x86/kernel/amd_iommu.c      |    4 ----
 arch/x86/kernel/amd_iommu_init.c |    9 +++++++++
 2 files changed, 9 insertions(+), 4 deletions(-)
arch/x86/kernel/amd_iommu.c +0 -4
index 8a9aaa8..0d20286 100644
@@ -2330,10 +2330,6 @@ int __init amd_iommu_init_dma_ops(void)
 
 	iommu_detected = 1;
 	swiotlb = 0;
-#ifdef CONFIG_GART_IOMMU
-	gart_iommu_aperture_disabled = 1;
-	gart_iommu_aperture = 0;
-#endif
 
 	/* Make the driver finally visible to the drivers */
 	dma_ops = &amd_iommu_dma_ops;
arch/x86/kernel/amd_iommu_init.c +9 -0
index 1405346..3cc63e2 100644
@@ -1358,6 +1358,15 @@ free:
 
 	free_unity_maps();
 
+#ifdef CONFIG_GART_IOMMU
+	/*
+	 * We failed to initialize the AMD IOMMU - try fallback to GART
+	 * if possible.
+	 */
+	gart_iommu_init();
+
+#endif
+
 	goto out;
 }
 
Signature · 1 line
1.7.1
#4 lore raw
In reply to Joerg Roedel
Quoted 17 lines · Joerg Roedel

Hi Ingo,

The following changes since commit 67a3e12b05e055c0415c556a315a3d3eb637e29e:

Linux 2.6.35-rc1 (2010-05-30 13:21:02 -0700)

are available in the git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/joro/linux-2.6-iommu.git amd-iommu/2.6.35

Joerg Roedel (3):
Merge commit 'v2.6.35-rc1' into amd-iommu/2.6.35 x86/amd-iommu: Fix crash when request_mem_region fails x86/amd-iommu: Fall back to GART if initialization fails

Julia Lawall (1):
arch/x86/kernel: Add missing spin_unlock

Quoted 8 lines
 arch/x86/kernel/amd_iommu.c      |   16 +++++++++-------
 arch/x86/kernel/amd_iommu_init.c |   20 +++++++++++++++++---
 2 files changed, 26 insertions(+), 10 deletions(-)

These patches fix two bugs in the error-paths of the amd iommu driver. One fix adds missing spin_unlocks and the other two patches fix a problem with broken BIOSes where the driver can't allocate the mmio region for the iommu hardware. Please pull.

Pulled, thanks Joerg!

Ingo