Skip to contentSkip to conversation

IOMMU updates for 2.6.32-rc5

GIT PULL In review Joerg Roedel 2 participants 6 messages Oct 29, 2009 – Nov 5, 2009

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

Hi Ingo,

The following changes since commit 012abeea669ea49636cf952d13298bb68654146a:
Linus Torvalds (1):

        Linux 2.6.32-rc5

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/joro/linux-2.6-iommu.git iommu/fixes

Joerg Roedel (4):
x86/amd-iommu: Workaround for erratum 63
x86/amd-iommu: Un__init function required on shutdown
dma-debug: Fix compile warning with PAE enabled
Merge branches 'amd-iommu/fixes', 'dma-debug/fixes' and 'gart/fixes' into iommu/fixes

Pavel Vasilyev (1):
agp/amd64: Remove GART dependency on AGP_AMD64

 arch/x86/include/asm/amd_iommu.h |    1 +
 arch/x86/kernel/amd_iommu.c      |    2 ++
 arch/x86/kernel/amd_iommu_init.c |   24 +++++++++++++++++++++++-
 drivers/char/agp/Kconfig         |    3 +--
 lib/dma-debug.c                  |    6 ++++--
 5 files changed, 31 insertions(+), 5 deletions(-)

The patches contain two important fixes for the AMD IOMMU driver (one erratum workaround and a section mismatch that only shows when the compiler does not inline the iommu_feature_disable function). Also included in this pull request is da dma-debug fix for a compilation warning on 32bit with PAE enabled and a previously missed patch which removes the build dependency between GART and AGP_AMD64 code. Please pull.

Joerg

#1 lore raw

PATCH 1/4 x86/amd-iommu: Workaround for erratum 63

There is an erratum for IOMMU hardware which documents undefined behavior when forwarding SMI requests from peripherals and the DTE of that peripheral has a sysmgt value of 01b. This problem caused weird IO_PAGE_FAULTS in my case.
This patch implements the suggested workaround for that erratum into the AMD IOMMU driver. The erratum is documented with number 63.

---
 arch/x86/include/asm/amd_iommu.h |    1 +
 arch/x86/kernel/amd_iommu.c      |    2 ++
 arch/x86/kernel/amd_iommu_init.c |   22 ++++++++++++++++++++++
 3 files changed, 25 insertions(+), 0 deletions(-)
arch/x86/include/asm/amd_iommu.h +1 -0
index bdf96f1..9dbd403 100644
@@ -30,6 +30,7 @@ extern irqreturn_t amd_iommu_int_handler(int irq, void *data);
 extern void amd_iommu_flush_all_domains(void);
 extern void amd_iommu_flush_all_devices(void);
 extern void amd_iommu_shutdown(void);
+extern void amd_iommu_apply_erratum_63(u16 devid);
 #else
 static inline int amd_iommu_init(void) { return -ENODEV; }
 static inline void amd_iommu_detect(void) { }
arch/x86/kernel/amd_iommu.c +2 -0
index 6c99f50..f95dfe5 100644
@@ -1114,6 +1114,8 @@ static void __detach_device(struct protection_domain *domain, u16 devid)
 	amd_iommu_dev_table[devid].data[1] = 0;
 	amd_iommu_dev_table[devid].data[2] = 0;
 
+	amd_iommu_apply_erratum_63(devid);
+
 	/* decrease reference counter */
 	domain->dev_cnt -= 1;
 
arch/x86/kernel/amd_iommu_init.c +22 -0
index c1b17e9..498c8c7 100644
@@ -509,6 +509,26 @@ static void set_dev_entry_bit(u16 devid, u8 bit)
 	amd_iommu_dev_table[devid].data[i] |= (1 << _bit);
 }
 
+static int get_dev_entry_bit(u16 devid, u8 bit)
+{
+	int i = (bit >> 5) & 0x07;
+	int _bit = bit & 0x1f;
+
+	return (amd_iommu_dev_table[devid].data[i] & (1 << _bit)) >> _bit;
+}
+
+
+void amd_iommu_apply_erratum_63(u16 devid)
+{
+	int sysmgt;
+
+	sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
+		 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
+
+	if (sysmgt == 0x01)
+		set_dev_entry_bit(devid, DEV_ENTRY_IW);
+}
+
 /* Writes the specific IOMMU for a device into the rlookup table */
 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
 {
@@ -537,6 +557,8 @@ static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
 	if (flags & ACPI_DEVFLAG_LINT1)
 		set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
 
+	amd_iommu_apply_erratum_63(devid);
+
 	set_iommu_for_device(iommu, devid);
 }
 
Signature · 1 line
1.6.5
#2 lore raw

Re: [PATCH 1/4] x86/amd-iommu: Workaround for erratum 63

Quoted 5 lines · Joerg Roedel

There is an erratum for IOMMU hardware which documents undefined behavior when forwarding SMI requests from peripherals and the DTE of that peripheral has a sysmgt value of 01b. This problem caused weird IO_PAGE_FAULTS in my case.

This should go to comment.

arch/x86/kernel/amd_iommu_init.c static void set_dev_entry_bit(u16 devid, u8 bit) · line 524

Quoted 6 lines
arch/x86/kernel/amd_iommu_init.c +6 -0
+
+
+void amd_iommu_apply_erratum_63(u16 devid)
+{
+	int sysmgt;
+

...that should go here. Pointer to pdf with erratum would be cool, too.
Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#3 lore raw

PATCH 2/4 x86/amd-iommu: Un__init function required on shutdown

In reply to Joerg Roedel

The function iommu_feature_disable is required on system shutdown to disable the IOMMU but it is marked as __init. This may result in a panic if the memory is reused. This patch fixes this bug.

---
 arch/x86/kernel/amd_iommu_init.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/x86/kernel/amd_iommu_init.c +1 -1
index 498c8c7..1e423b2 100644
@@ -240,7 +240,7 @@ static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
 	writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
 }
 
-static void __init iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
+static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
 {
 	u32 ctrl;
 
Signature · 1 line
1.6.5
#4 lore raw

PATCH 3/4 dma-debug: Fix compile warning with PAE enabled

In reply to Joerg Roedel

When PAE is enabled in the kernel configuration the size of phys_addr_t differs from the size of a void pointer. The gcc prints a warning about that in dma-debug code. This patch fixes the warning by converting the output to unsigned long long instead of a pointer.

---
 lib/dma-debug.c |    6 ++++--
 1 files changed, 4 insertions(+), 2 deletions(-)
lib/dma-debug.c +4 -2
index 58a9f9f..ce6b7ea 100644
@@ -819,9 +819,11 @@ static void check_unmap(struct dma_debug_entry *ref)
 		err_printk(ref->dev, entry, "DMA-API: device driver frees "
 			   "DMA memory with different CPU address "
 			   "[device address=0x%016llx] [size=%llu bytes] "
-			   "[cpu alloc address=%p] [cpu free address=%p]",
+			   "[cpu alloc address=0x%016llx] "
+			   "[cpu free address=0x%016llx]",
 			   ref->dev_addr, ref->size,
-			   (void *)entry->paddr, (void *)ref->paddr);
+			   (unsigned long long)entry->paddr,
+			   (unsigned long long)ref->paddr);
 	}
 
 	if (ref->sg_call_ents && ref->type == dma_debug_sg &&
Signature · 1 line
1.6.5
#5 lore raw

PATCH 4/4 agp/amd64: Remove GART dependency on AGP_AMD64

In reply to Joerg Roedel

From: Pavel Vasilyev <[email protected]>

The GART IOMMU code has no strong dependency to the AMD64 AGP code. So the automatic selection of AGP_AMD64 for GART can be removed.

---
 drivers/char/agp/Kconfig |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)
drivers/char/agp/Kconfig +1 -2
index ccb1fa8..2fb3a48 100644
@@ -56,9 +56,8 @@ config AGP_AMD
 	  X on AMD Irongate, 761, and 762 chipsets.
 
 config AGP_AMD64
-	tristate "AMD Opteron/Athlon64 on-CPU GART support" if !GART_IOMMU
+	tristate "AMD Opteron/Athlon64 on-CPU GART support"
 	depends on AGP && X86
-	default y if GART_IOMMU
 	help
 	  This option gives you AGP support for the GLX component of
 	  X using the on-CPU northbridge of the AMD Athlon64/Opteron CPUs.
Signature · 1 line
1.6.5