[git pull] IOMMU updates for 2.6.33

[git pull] IOMMU updates for 2.6.33

[Joerg Roedel]

Hi Ingo,


The following changes since commit 273bee27fa9f79d94b78c83506016f2e41e78983:
  FUJITA Tomonori (1):
        x86: Fix iommu=soft boot option

are available in the git repository at:

  git://git.kernel.org/pub/scm/linux/kernel/git/joro/linux-2.6-iommu.git amd-iommu/2.6.33

Joerg Roedel (35):
      x86/amd-iommu: Separate internal interface definitions
      x86/amd-iommu: Update copyright headers
      x86/amd-iommu: Add an index field to struct amd_iommu
      x86/amd-iommu: Add per IOMMU reference counting
      x86/amd-iommu: Add function to complete a tlb flush
      x86/amd-iommu: Make iommu_flush_pages aware of multiple IOMMUs
      x86/amd-iommu: Use __iommu_flush_pages for tlb flushes
      x86/amd-iommu: Remove iommu_flush_domain function
      x86/amd-iommu: Implement protection domain list
      x86/amd-iommu: Reimplement amd_iommu_flush_all_domains()
      x86/amd-iommu: Reimplement flush_all_domains_on_iommu()
      x86/amd-iommu: Make np-cache a global flag
      x86/amd-iommu: Use check_device for amd_iommu_dma_supported
      x86/amd-iommu: Use check_device in get_device_resources
      x86/amd-iommu: Remove iommu parameter from dma_ops_domain_(un)map
      x86/amd-iommu: Make alloc_new_range aware of multiple IOMMUs
      x86/amd-iommu: Remove iommu parameter from __(un)map_single
      x86/amd-iommu: Remove iommu specific handling from dma_ops path
      x86/amd-iommu: Let domain_for_device handle aliases
      x86/amd-iommu: Simplify get_device_resources()
      x86/amd-iommu: Move find_protection_domain to helper functions
      x86/amd-iommu: Use get_device_id and check_device where appropriate
      x86/amd-iommu: Remove iommu parameter from dma_ops_domain_alloc
      x86/amd-iommu: Move some pte allocation functions in the right section
      x86/amd-iommu: Rearrange dma_ops related functions
      x86/amd-iommu: Remove support for domain sharing
      x86/amd-iommu: Use dev->arch->iommu to store iommu related information
      x86/amd-iommu: Add device bind reference counting
      x86/amd-iommu: Keep devices per domain in a list
      x86/amd-iommu: Cleanup attach/detach_device code
      x86/amd-iommu: Introduce iommu_flush_device() function
      x86/amd-iommu: Cleanup DTE flushing code
      x86/amd-iommu: Move reset_iommu_command_buffer out of locked code
      x86/amd-iommu: Remove amd_iommu_pd_table
      Merge branch 'gart/fixes' into amd-iommu/2.6.33

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

 arch/x86/include/asm/amd_iommu.h       |   12 +-
 arch/x86/include/asm/amd_iommu_proto.h |   38 +
 arch/x86/include/asm/amd_iommu_types.h |   54 ++-
 arch/x86/include/asm/device.h          |    2 +-
 arch/x86/kernel/amd_iommu.c            | 1230 +++++++++++++++++---------------
 arch/x86/kernel/amd_iommu_init.c       |   67 +-
 drivers/char/agp/Kconfig               |    3 +-
 7 files changed, 777 insertions(+), 629 deletions(-)
 create mode 100644 arch/x86/include/asm/amd_iommu_proto.h

Most of the changes are for AMD IOMMU driver. The changes there are bigger than
usual because I did a lot of changes to the data structures storing the
relations between devices, protection domains and iommus. A major change is
that the amd_iommu_pd_table is now gone and replaced by a struct iommu_dev_data
for each device stored in dev->archdata->iommu.
These changes result in a more robust reference counting, especially for
devices with an alias and simplified code in many places.
Also included are some related cleanups which separated the internal function
prototypes to a several header and code movement to get rid of most function
forward declarations. Please pull.

	Joerg

[PATCH 01/34] x86/amd-iommu: Separate internal interface definitions

[Joerg Roedel]

This patch moves all function declarations which are only
used inside the driver code to a seperate header file.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu.h       |   10 +++-----
 arch/x86/include/asm/amd_iommu_proto.h |   38 ++++++++++++++++++++++++++++++++
 arch/x86/include/asm/amd_iommu_types.h |    5 ----
 arch/x86/kernel/amd_iommu.c            |    1 +
 arch/x86/kernel/amd_iommu_init.c       |    1 +
 5 files changed, 44 insertions(+), 11 deletions(-)
 create mode 100644 arch/x86/include/asm/amd_iommu_proto.h

diff --git a/arch/x86/include/asm/amd_iommu.h b/arch/x86/include/asm/amd_iommu.h
index b8ef2ee..0891338 100644
--- a/arch/x86/include/asm/amd_iommu.h
+++ b/arch/x86/include/asm/amd_iommu.h
@@ -23,15 +23,13 @@
 #include <linux/irqreturn.h>
 
 #ifdef CONFIG_AMD_IOMMU
-extern int amd_iommu_init_dma_ops(void);
-extern int amd_iommu_init_passthrough(void);
+
 extern void amd_iommu_detect(void);
-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_apply_erratum_63(u16 devid);
+
 #else
+
 static inline void amd_iommu_detect(void) { }
+
 #endif
 
 #endif /* _ASM_X86_AMD_IOMMU_H */
diff --git a/arch/x86/include/asm/amd_iommu_proto.h b/arch/x86/include/asm/amd_iommu_proto.h
new file mode 100644
index 0000000..84786fb
--- /dev/null
+++ b/arch/x86/include/asm/amd_iommu_proto.h
@@ -0,0 +1,38 @@
+/*
+ * Copyright (C) 2009 Advanced Micro Devices, Inc.
+ * Author: Joerg Roedel <joerg.roedel@amd.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
+ */
+
+#ifndef _ASM_X86_AMD_IOMMU_PROTO_H
+#define _ASM_X86_AMD_IOMMU_PROTO_H
+
+struct amd_iommu;
+
+extern int amd_iommu_init_dma_ops(void);
+extern int amd_iommu_init_passthrough(void);
+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_apply_erratum_63(u16 devid);
+extern void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu);
+
+#ifndef CONFIG_AMD_IOMMU_STATS
+
+static inline void amd_iommu_stats_init(void) { }
+
+#endif /* !CONFIG_AMD_IOMMU_STATS */
+
+#endif /* _ASM_X86_AMD_IOMMU_PROTO_H  */
diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index 2a2cc7a..27db7f9 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -462,11 +462,6 @@ struct __iommu_counter {
 #define ADD_STATS_COUNTER(name, x)
 #define SUB_STATS_COUNTER(name, x)
 
-static inline void amd_iommu_stats_init(void) { }
-
 #endif /* CONFIG_AMD_IOMMU_STATS */
 
-/* some function prototypes */
-extern void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu);
-
 #endif /* _ASM_X86_AMD_IOMMU_TYPES_H */
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index b74b212..50d2b05 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -28,6 +28,7 @@
 #include <asm/proto.h>
 #include <asm/iommu.h>
 #include <asm/gart.h>
+#include <asm/amd_iommu_proto.h>
 #include <asm/amd_iommu_types.h>
 #include <asm/amd_iommu.h>
 
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index 72bdbda..db30cfe 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -25,6 +25,7 @@
 #include <linux/interrupt.h>
 #include <linux/msi.h>
 #include <asm/pci-direct.h>
+#include <asm/amd_iommu_proto.h>
 #include <asm/amd_iommu_types.h>
 #include <asm/amd_iommu.h>
 #include <asm/iommu.h>
-- 
1.6.5.3

[PATCH 02/34] x86/amd-iommu: Update copyright headers

[Joerg Roedel]

This patch updates the copyright headers in the relevant AMD
IOMMU driver files to match the date of the latest changes.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu.h       |    2 +-
 arch/x86/include/asm/amd_iommu_types.h |    2 +-
 arch/x86/kernel/amd_iommu.c            |    2 +-
 arch/x86/kernel/amd_iommu_init.c       |    2 +-
 4 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/arch/x86/include/asm/amd_iommu.h b/arch/x86/include/asm/amd_iommu.h
index 0891338..5af2982 100644
--- a/arch/x86/include/asm/amd_iommu.h
+++ b/arch/x86/include/asm/amd_iommu.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
+ * Copyright (C) 2007-2009 Advanced Micro Devices, Inc.
  * Author: Joerg Roedel <joerg.roedel@amd.com>
  *         Leo Duran <leo.duran@amd.com>
  *
diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index 27db7f9..df5e9c8 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
+ * Copyright (C) 2007-2009 Advanced Micro Devices, Inc.
  * Author: Joerg Roedel <joerg.roedel@amd.com>
  *         Leo Duran <leo.duran@amd.com>
  *
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 50d2b05..7fe28be 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
+ * Copyright (C) 2007-2009 Advanced Micro Devices, Inc.
  * Author: Joerg Roedel <joerg.roedel@amd.com>
  *         Leo Duran <leo.duran@amd.com>
  *
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index db30cfe..cee1142 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
+ * Copyright (C) 2007-2009 Advanced Micro Devices, Inc.
  * Author: Joerg Roedel <joerg.roedel@amd.com>
  *         Leo Duran <leo.duran@amd.com>
  *
-- 
1.6.5.3

[PATCH 03/34] x86/amd-iommu: Add an index field to struct amd_iommu

[Joerg Roedel]

This patch adds an index field to struct amd_iommu which can
be used to lookup it up in an array. This index will be used
in struct protection_domain to keep track which protection
domain has devices behind which IOMMU.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu_types.h |   17 +++++++++++++++++
 arch/x86/kernel/amd_iommu_init.c       |   15 +++++++++++++++
 2 files changed, 32 insertions(+), 0 deletions(-)

diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index df5e9c8..ab3e7bf 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -25,6 +25,11 @@
 #include <linux/spinlock.h>
 
 /*
+ * Maximum number of IOMMUs supported
+ */
+#define MAX_IOMMUS	32
+
+/*
  * some size calculation constants
  */
 #define DEV_TABLE_ENTRY_SIZE		32
@@ -291,6 +296,9 @@ struct dma_ops_domain {
 struct amd_iommu {
 	struct list_head list;
 
+	/* Index within the IOMMU array */
+	int index;
+
 	/* locks the accesses to the hardware */
 	spinlock_t lock;
 
@@ -357,6 +365,15 @@ struct amd_iommu {
 extern struct list_head amd_iommu_list;
 
 /*
+ * Array with pointers to each IOMMU struct
+ * The indices are referenced in the protection domains
+ */
+extern struct amd_iommu *amd_iommus[MAX_IOMMUS];
+
+/* Number of IOMMUs present in the system */
+extern int amd_iommus_present;
+
+/*
  * Structure defining one entry in the device table
  */
 struct dev_table_entry {
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index cee1142..8567d16 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -137,6 +137,10 @@ bool amd_iommu_unmap_flush;		/* if true, flush on every unmap */
 LIST_HEAD(amd_iommu_list);		/* list of all AMD IOMMUs in the
 					   system */
 
+/* Array to assign indices to IOMMUs*/
+struct amd_iommu *amd_iommus[MAX_IOMMUS];
+int amd_iommus_present;
+
 /*
  * Pointer to the device table which is shared by all AMD IOMMUs
  * it is indexed by the PCI device id or the HT unit id and contains
@@ -840,7 +844,18 @@ static void __init free_iommu_all(void)
 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
 {
 	spin_lock_init(&iommu->lock);
+
+	/* Add IOMMU to internal data structures */
 	list_add_tail(&iommu->list, &amd_iommu_list);
+	iommu->index             = amd_iommus_present++;
+
+	if (unlikely(iommu->index >= MAX_IOMMUS)) {
+		WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
+		return -ENOSYS;
+	}
+
+	/* Index is fine - add IOMMU to the array */
+	amd_iommus[iommu->index] = iommu;
 
 	/*
 	 * Copy data from ACPI table entry to the iommu struct
-- 
1.6.5.3

[PATCH 04/34] x86/amd-iommu: Add per IOMMU reference counting

[Joerg Roedel]

This patch adds reference counting for protection domains
per IOMMU. This allows a smarter TLB flushing strategy.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu_types.h |    2 ++
 arch/x86/kernel/amd_iommu.c            |   12 +++++++++---
 2 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index ab3e7bf..e68b148 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -238,7 +238,9 @@ struct protection_domain {
 	unsigned long flags;	/* flags to find out type of domain */
 	bool updated;		/* complete domain flush required */
 	unsigned dev_cnt;	/* devices assigned to this domain */
+	unsigned dev_iommu[MAX_IOMMUS]; /* per-IOMMU reference count */
 	void *priv;		/* private data */
+
 };
 
 /*
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 7fe28be..8c38f00 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1175,7 +1175,9 @@ static void __attach_device(struct amd_iommu *iommu,
 	/* update DTE entry */
 	set_dte_entry(devid, domain);
 
-	domain->dev_cnt += 1;
+	/* Do reference counting */
+	domain->dev_iommu[iommu->index] += 1;
+	domain->dev_cnt                 += 1;
 
 	/* ready */
 	spin_unlock(&domain->lock);
@@ -1209,6 +1211,9 @@ static void attach_device(struct amd_iommu *iommu,
  */
 static void __detach_device(struct protection_domain *domain, u16 devid)
 {
+	struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
+
+	BUG_ON(!iommu);
 
 	/* lock domain */
 	spin_lock(&domain->lock);
@@ -1223,8 +1228,9 @@ static void __detach_device(struct protection_domain *domain, u16 devid)
 
 	amd_iommu_apply_erratum_63(devid);
 
-	/* decrease reference counter */
-	domain->dev_cnt -= 1;
+	/* decrease reference counters */
+	domain->dev_iommu[iommu->index] -= 1;
+	domain->dev_cnt                 -= 1;
 
 	/* ready */
 	spin_unlock(&domain->lock);
-- 
1.6.5.3

[PATCH 05/34] x86/amd-iommu: Add function to complete a tlb flush

[Joerg Roedel]

This patch adds a function to the AMD IOMMU driver which
completes all queued commands an all IOMMUs a specific
domain has devices attached on. This is required in a later
patch when per-domain flushing is implemented.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   28 ++++++++++++++++++++++------
 1 files changed, 22 insertions(+), 6 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 8c38f00..8fa5cc3 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -376,6 +376,22 @@ out:
 	return 0;
 }
 
+static void iommu_flush_complete(struct protection_domain *domain)
+{
+	int i;
+
+	for (i = 0; i < amd_iommus_present; ++i) {
+		if (!domain->dev_iommu[i])
+			continue;
+
+		/*
+		 * Devices of this domain are behind this IOMMU
+		 * We need to wait for completion of all commands.
+		 */
+		iommu_completion_wait(amd_iommus[i]);
+	}
+}
+
 /*
  * Command send function for invalidating a device table entry
  */
@@ -1758,7 +1774,7 @@ static dma_addr_t map_page(struct device *dev, struct page *page,
 	if (addr == DMA_ERROR_CODE)
 		goto out;
 
-	iommu_completion_wait(iommu);
+	iommu_flush_complete(domain);
 
 out:
 	spin_unlock_irqrestore(&domain->lock, flags);
@@ -1791,7 +1807,7 @@ static void unmap_page(struct device *dev, dma_addr_t dma_addr, size_t size,
 
 	__unmap_single(iommu, domain->priv, dma_addr, size, dir);
 
-	iommu_completion_wait(iommu);
+	iommu_flush_complete(domain);
 
 	spin_unlock_irqrestore(&domain->lock, flags);
 }
@@ -1863,7 +1879,7 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
 			goto unmap;
 	}
 
-	iommu_completion_wait(iommu);
+	iommu_flush_complete(domain);
 
 out:
 	spin_unlock_irqrestore(&domain->lock, flags);
@@ -1914,7 +1930,7 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
 		s->dma_address = s->dma_length = 0;
 	}
 
-	iommu_completion_wait(iommu);
+	iommu_flush_complete(domain);
 
 	spin_unlock_irqrestore(&domain->lock, flags);
 }
@@ -1969,7 +1985,7 @@ static void *alloc_coherent(struct device *dev, size_t size,
 		goto out_free;
 	}
 
-	iommu_completion_wait(iommu);
+	iommu_flush_complete(domain);
 
 	spin_unlock_irqrestore(&domain->lock, flags);
 
@@ -2010,7 +2026,7 @@ static void free_coherent(struct device *dev, size_t size,
 
 	__unmap_single(iommu, domain->priv, dma_addr, size, DMA_BIDIRECTIONAL);
 
-	iommu_completion_wait(iommu);
+	iommu_flush_complete(domain);
 
 	spin_unlock_irqrestore(&domain->lock, flags);
 
-- 
1.6.5.3

[PATCH 06/34] x86/amd-iommu: Make iommu_flush_pages aware of multiple IOMMUs

[Joerg Roedel]

This patch extends the iommu_flush_pages function to flush
the TLB entries on all IOMMUs the domain has devices on.
This basically gives up the former assumption that dma_ops
domains are only bound to one IOMMU in the system.
For dma_ops domains this is still true but not for
IOMMU-API managed domains. Giving this assumption up for
dma_ops domains too allows code simplification.
Further it splits out the main logic into a generic function
which can be used by iommu_flush_tlb too.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   31 ++++++++++++++++++++++++-------
 1 files changed, 24 insertions(+), 7 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 8fa5cc3..7c06e57 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -447,10 +447,10 @@ static int iommu_queue_inv_iommu_pages(struct amd_iommu *iommu,
  * It invalidates a single PTE if the range to flush is within a single
  * page. Otherwise it flushes the whole TLB of the IOMMU.
  */
-static int iommu_flush_pages(struct amd_iommu *iommu, u16 domid,
-		u64 address, size_t size)
+static void __iommu_flush_pages(struct protection_domain *domain,
+				u64 address, size_t size, int pde)
 {
-	int s = 0;
+	int s = 0, i;
 	unsigned pages = iommu_num_pages(address, size, PAGE_SIZE);
 
 	address &= PAGE_MASK;
@@ -464,9 +464,26 @@ static int iommu_flush_pages(struct amd_iommu *iommu, u16 domid,
 		s = 1;
 	}
 
-	iommu_queue_inv_iommu_pages(iommu, address, domid, 0, s);
 
-	return 0;
+	for (i = 0; i < amd_iommus_present; ++i) {
+		if (!domain->dev_iommu[i])
+			continue;
+
+		/*
+		 * Devices of this domain are behind this IOMMU
+		 * We need a TLB flush
+		 */
+		iommu_queue_inv_iommu_pages(amd_iommus[i], address,
+					    domain->id, pde, s);
+	}
+
+	return;
+}
+
+static void iommu_flush_pages(struct protection_domain *domain,
+			     u64 address, size_t size)
+{
+	__iommu_flush_pages(domain, address, size, 0);
 }
 
 /* Flush the whole IO/TLB for a given protection domain */
@@ -1683,7 +1700,7 @@ retry:
 		iommu_flush_tlb(iommu, dma_dom->domain.id);
 		dma_dom->need_flush = false;
 	} else if (unlikely(iommu_has_npcache(iommu)))
-		iommu_flush_pages(iommu, dma_dom->domain.id, address, size);
+		iommu_flush_pages(&dma_dom->domain, address, size);
 
 out:
 	return address;
@@ -1731,7 +1748,7 @@ static void __unmap_single(struct amd_iommu *iommu,
 	dma_ops_free_addresses(dma_dom, dma_addr, pages);
 
 	if (amd_iommu_unmap_flush || dma_dom->need_flush) {
-		iommu_flush_pages(iommu, dma_dom->domain.id, dma_addr, size);
+		iommu_flush_pages(&dma_dom->domain, dma_addr, size);
 		dma_dom->need_flush = false;
 	}
 }
-- 
1.6.5.3

[PATCH 07/34] x86/amd-iommu: Use __iommu_flush_pages for tlb flushes

[Joerg Roedel]

This patch re-implements iommu_flush_tlb functions to use
the __iommu_flush_pages logic.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   22 +++++++---------------
 1 files changed, 7 insertions(+), 15 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 7c06e57..c55aa07 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -451,7 +451,7 @@ static void __iommu_flush_pages(struct protection_domain *domain,
 				u64 address, size_t size, int pde)
 {
 	int s = 0, i;
-	unsigned pages = iommu_num_pages(address, size, PAGE_SIZE);
+	unsigned long pages = iommu_num_pages(address, size, PAGE_SIZE);
 
 	address &= PAGE_MASK;
 
@@ -487,23 +487,15 @@ static void iommu_flush_pages(struct protection_domain *domain,
 }
 
 /* Flush the whole IO/TLB for a given protection domain */
-static void iommu_flush_tlb(struct amd_iommu *iommu, u16 domid)
+static void iommu_flush_tlb(struct protection_domain *domain)
 {
-	u64 address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS;
-
-	INC_STATS_COUNTER(domain_flush_single);
-
-	iommu_queue_inv_iommu_pages(iommu, address, domid, 0, 1);
+	__iommu_flush_pages(domain, 0, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, 0);
 }
 
 /* Flush the whole IO/TLB for a given protection domain - including PDE */
-static void iommu_flush_tlb_pde(struct amd_iommu *iommu, u16 domid)
+static void iommu_flush_tlb_pde(struct protection_domain *domain)
 {
-       u64 address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS;
-
-       INC_STATS_COUNTER(domain_flush_single);
-
-       iommu_queue_inv_iommu_pages(iommu, address, domid, 1, 1);
+	__iommu_flush_pages(domain, 0, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, 1);
 }
 
 /*
@@ -1236,7 +1228,7 @@ static void attach_device(struct amd_iommu *iommu,
 	 * here to evict all dirty stuff.
 	 */
 	iommu_queue_inv_dev_entry(iommu, devid);
-	iommu_flush_tlb_pde(iommu, domain->id);
+	iommu_flush_tlb_pde(domain);
 }
 
 /*
@@ -1697,7 +1689,7 @@ retry:
 	ADD_STATS_COUNTER(alloced_io_mem, size);
 
 	if (unlikely(dma_dom->need_flush && !amd_iommu_unmap_flush)) {
-		iommu_flush_tlb(iommu, dma_dom->domain.id);
+		iommu_flush_tlb(&dma_dom->domain);
 		dma_dom->need_flush = false;
 	} else if (unlikely(iommu_has_npcache(iommu)))
 		iommu_flush_pages(&dma_dom->domain, address, size);
-- 
1.6.5.3

[PATCH 08/34] x86/amd-iommu: Remove iommu_flush_domain function

[Joerg Roedel]

This iommu_flush_tlb_pde function does essentially the same.
So the iommu_flush_domain function is redundant and can be
removed.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   18 ++----------------
 1 files changed, 2 insertions(+), 16 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index c55aa07..b2c19f4 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -528,20 +528,6 @@ static void flush_all_domains_on_iommu(struct amd_iommu *iommu)
 
 }
 
-/*
- * This function is used to flush the IO/TLB for a given protection domain
- * on every IOMMU in the system
- */
-static void iommu_flush_domain(u16 domid)
-{
-	struct amd_iommu *iommu;
-
-	INC_STATS_COUNTER(domain_flush_all);
-
-	for_each_iommu(iommu)
-		flush_domain_on_iommu(iommu, domid);
-}
-
 void amd_iommu_flush_all_domains(void)
 {
 	struct amd_iommu *iommu;
@@ -1464,7 +1450,7 @@ static void update_domain(struct protection_domain *domain)
 
 	update_device_table(domain);
 	flush_devices_by_domain(domain);
-	iommu_flush_domain(domain->id);
+	iommu_flush_tlb_pde(domain);
 
 	domain->updated = false;
 }
@@ -2377,7 +2363,7 @@ static void amd_iommu_unmap_range(struct iommu_domain *dom,
 		iova  += PAGE_SIZE;
 	}
 
-	iommu_flush_domain(domain->id);
+	iommu_flush_tlb_pde(domain);
 }
 
 static phys_addr_t amd_iommu_iova_to_phys(struct iommu_domain *dom,
-- 
1.6.5.3

[PATCH 09/34] x86/amd-iommu: Implement protection domain list

[Joerg Roedel]

This patch adds code to keep a global list of all protection
domains. This allows to simplify the resume code.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu_types.h |    7 ++++++
 arch/x86/kernel/amd_iommu.c            |   33 ++++++++++++++++++++++++++++++++
 arch/x86/kernel/amd_iommu_init.c       |    8 +++++++
 3 files changed, 48 insertions(+), 0 deletions(-)

diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index e68b148..b332b7f 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -231,6 +231,7 @@ extern bool amd_iommu_dump;
  * independent of their use.
  */
 struct protection_domain {
+	struct list_head list;  /* for list of all protection domains */
 	spinlock_t lock;	/* mostly used to lock the page table*/
 	u16 id;			/* the domain id written to the device table */
 	int mode;		/* paging mode (0-6 levels) */
@@ -376,6 +377,12 @@ extern struct amd_iommu *amd_iommus[MAX_IOMMUS];
 extern int amd_iommus_present;
 
 /*
+ * Declarations for the global list of all protection domains
+ */
+extern spinlock_t amd_iommu_pd_lock;
+extern struct list_head amd_iommu_pd_list;
+
+/*
  * Structure defining one entry in the device table
  */
 struct dev_table_entry {
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index b2c19f4..0c4319b 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -985,6 +985,31 @@ static void dma_ops_free_addresses(struct dma_ops_domain *dom,
  *
  ****************************************************************************/
 
+/*
+ * This function adds a protection domain to the global protection domain list
+ */
+static void add_domain_to_list(struct protection_domain *domain)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&amd_iommu_pd_lock, flags);
+	list_add(&domain->list, &amd_iommu_pd_list);
+	spin_unlock_irqrestore(&amd_iommu_pd_lock, flags);
+}
+
+/*
+ * This function removes a protection domain to the global
+ * protection domain list
+ */
+static void del_domain_from_list(struct protection_domain *domain)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&amd_iommu_pd_lock, flags);
+	list_del(&domain->list);
+	spin_unlock_irqrestore(&amd_iommu_pd_lock, flags);
+}
+
 static u16 domain_id_alloc(void)
 {
 	unsigned long flags;
@@ -1073,6 +1098,8 @@ static void dma_ops_domain_free(struct dma_ops_domain *dom)
 	if (!dom)
 		return;
 
+	del_domain_from_list(&dom->domain);
+
 	free_pagetable(&dom->domain);
 
 	for (i = 0; i < APERTURE_MAX_RANGES; ++i) {
@@ -1113,6 +1140,8 @@ static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu)
 	dma_dom->need_flush = false;
 	dma_dom->target_dev = 0xffff;
 
+	add_domain_to_list(&dma_dom->domain);
+
 	if (alloc_new_range(iommu, dma_dom, true, GFP_KERNEL))
 		goto free_dma_dom;
 
@@ -2188,6 +2217,8 @@ static void protection_domain_free(struct protection_domain *domain)
 	if (!domain)
 		return;
 
+	del_domain_from_list(domain);
+
 	if (domain->id)
 		domain_id_free(domain->id);
 
@@ -2207,6 +2238,8 @@ static struct protection_domain *protection_domain_alloc(void)
 	if (!domain->id)
 		goto out_err;
 
+	add_domain_to_list(domain);
+
 	return domain;
 
 out_err:
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index 8567d16..73d5173 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -142,6 +142,12 @@ struct amd_iommu *amd_iommus[MAX_IOMMUS];
 int amd_iommus_present;
 
 /*
+ * List of protection domains - used during resume
+ */
+LIST_HEAD(amd_iommu_pd_list);
+spinlock_t amd_iommu_pd_lock;
+
+/*
  * Pointer to the device table which is shared by all AMD IOMMUs
  * it is indexed by the PCI device id or the HT unit id and contains
  * information about the domain the device belongs to as well as the
@@ -1263,6 +1269,8 @@ static int __init amd_iommu_init(void)
 	 */
 	amd_iommu_pd_alloc_bitmap[0] = 1;
 
+	spin_lock_init(&amd_iommu_pd_lock);
+
 	/*
 	 * now the data structures are allocated and basically initialized
 	 * start the real acpi table scan
-- 
1.6.5.3

[PATCH 10/34] x86/amd-iommu: Reimplement amd_iommu_flush_all_domains()

[Joerg Roedel]

This patch reimplementes the amd_iommu_flush_all_domains
function to use the global protection domain list instead
of flushing every domain on every IOMMU.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |    8 +++++---
 1 files changed, 5 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 0c4319b..5141f56 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -530,10 +530,12 @@ static void flush_all_domains_on_iommu(struct amd_iommu *iommu)
 
 void amd_iommu_flush_all_domains(void)
 {
-	struct amd_iommu *iommu;
+	struct protection_domain *domain;
 
-	for_each_iommu(iommu)
-		flush_all_domains_on_iommu(iommu);
+	list_for_each_entry(domain, &amd_iommu_pd_list, list) {
+		iommu_flush_tlb_pde(domain);
+		iommu_flush_complete(domain);
+	}
 }
 
 static void flush_all_devices_for_iommu(struct amd_iommu *iommu)
-- 
1.6.5.3

[PATCH 11/34] x86/amd-iommu: Reimplement flush_all_domains_on_iommu()

[Joerg Roedel]

This patch reimplements the function
flush_all_domains_on_iommu to use the global protection
domain list.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   43 ++++++++++++++++++++++++-------------------
 1 files changed, 24 insertions(+), 19 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 5141f56..a1bd99d 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -499,43 +499,48 @@ static void iommu_flush_tlb_pde(struct protection_domain *domain)
 }
 
 /*
- * This function flushes one domain on one IOMMU
+ * This function flushes all domains that have devices on the given IOMMU
  */
-static void flush_domain_on_iommu(struct amd_iommu *iommu, u16 domid)
+static void flush_all_domains_on_iommu(struct amd_iommu *iommu)
 {
-	struct iommu_cmd cmd;
+	u64 address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS;
+	struct protection_domain *domain;
 	unsigned long flags;
 
-	__iommu_build_inv_iommu_pages(&cmd, CMD_INV_IOMMU_ALL_PAGES_ADDRESS,
-				      domid, 1, 1);
-
-	spin_lock_irqsave(&iommu->lock, flags);
-	__iommu_queue_command(iommu, &cmd);
-	__iommu_completion_wait(iommu);
-	__iommu_wait_for_completion(iommu);
-	spin_unlock_irqrestore(&iommu->lock, flags);
-}
-
-static void flush_all_domains_on_iommu(struct amd_iommu *iommu)
-{
-	int i;
+	spin_lock_irqsave(&amd_iommu_pd_lock, flags);
 
-	for (i = 1; i < MAX_DOMAIN_ID; ++i) {
-		if (!test_bit(i, amd_iommu_pd_alloc_bitmap))
+	list_for_each_entry(domain, &amd_iommu_pd_list, list) {
+		if (domain->dev_iommu[iommu->index] == 0)
 			continue;
-		flush_domain_on_iommu(iommu, i);
+
+		spin_lock(&domain->lock);
+		iommu_queue_inv_iommu_pages(iommu, address, domain->id, 1, 1);
+		iommu_flush_complete(domain);
+		spin_unlock(&domain->lock);
 	}
 
+	spin_unlock_irqrestore(&amd_iommu_pd_lock, flags);
 }
 
+/*
+ * This function uses heavy locking and may disable irqs for some time. But
+ * this is no issue because it is only called during resume.
+ */
 void amd_iommu_flush_all_domains(void)
 {
 	struct protection_domain *domain;
+	unsigned long flags;
+
+	spin_lock_irqsave(&amd_iommu_pd_lock, flags);
 
 	list_for_each_entry(domain, &amd_iommu_pd_list, list) {
+		spin_lock(&domain->lock);
 		iommu_flush_tlb_pde(domain);
 		iommu_flush_complete(domain);
+		spin_unlock(&domain->lock);
 	}
+
+	spin_unlock_irqrestore(&amd_iommu_pd_lock, flags);
 }
 
 static void flush_all_devices_for_iommu(struct amd_iommu *iommu)
-- 
1.6.5.3

[PATCH 12/34] x86/amd-iommu: Make np-cache a global flag

[Joerg Roedel]

The non-present cache flag was IOMMU local until now which
doesn't make sense. Make this a global flag so we can remove
the lase user of 'struct iommu' in the map/unmap path.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu_types.h |    3 +++
 arch/x86/kernel/amd_iommu.c            |    8 +-------
 arch/x86/kernel/amd_iommu_init.c       |    6 ++++++
 3 files changed, 10 insertions(+), 7 deletions(-)

diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index b332b7f..4899f78 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -211,6 +211,9 @@ extern bool amd_iommu_dump;
 			printk(KERN_INFO "AMD-Vi: " format, ## arg);	\
 	} while(0);
 
+/* global flag if IOMMUs cache non-present entries */
+extern bool amd_iommu_np_cache;
+
 /*
  * Make iterating over all IOMMUs easier
  */
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index a1bd99d..5ebd24e 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -131,12 +131,6 @@ static void amd_iommu_stats_init(void)
 
 #endif
 
-/* returns !0 if the IOMMU is caching non-present entries in its TLB */
-static int iommu_has_npcache(struct amd_iommu *iommu)
-{
-	return iommu->cap & (1UL << IOMMU_CAP_NPCACHE);
-}
-
 /****************************************************************************
  *
  * Interrupt handling functions
@@ -1713,7 +1707,7 @@ retry:
 	if (unlikely(dma_dom->need_flush && !amd_iommu_unmap_flush)) {
 		iommu_flush_tlb(&dma_dom->domain);
 		dma_dom->need_flush = false;
-	} else if (unlikely(iommu_has_npcache(iommu)))
+	} else if (unlikely(amd_iommu_np_cache))
 		iommu_flush_pages(&dma_dom->domain, address, size);
 
 out:
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index 73d5173..fbe4c3c 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -141,6 +141,9 @@ LIST_HEAD(amd_iommu_list);		/* list of all AMD IOMMUs in the
 struct amd_iommu *amd_iommus[MAX_IOMMUS];
 int amd_iommus_present;
 
+/* IOMMUs have a non-present cache? */
+bool amd_iommu_np_cache __read_mostly;
+
 /*
  * List of protection domains - used during resume
  */
@@ -891,6 +894,9 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
 	init_iommu_from_acpi(iommu, h);
 	init_iommu_devices(iommu);
 
+	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
+		amd_iommu_np_cache = true;
+
 	return pci_enable_device(iommu->dev);
 }
 
-- 
1.6.5.3

[PATCH 13/34] x86/amd-iommu: Use check_device for amd_iommu_dma_supported

[Joerg Roedel]

The check_device logic needs to include the dma_supported
checks to be really sure. Merge the dma_supported logic into
check_device and use it to implement dma_supported.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   35 +++++++++++++++++++----------------
 1 files changed, 19 insertions(+), 16 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 5ebd24e..ac27b1d 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1375,9 +1375,27 @@ static struct notifier_block device_nb = {
  */
 static bool check_device(struct device *dev)
 {
+	u16 bdf;
+	struct pci_dev *pcidev;
+
 	if (!dev || !dev->dma_mask)
 		return false;
 
+	/* No device or no PCI device */
+	if (!dev || dev->bus != &pci_bus_type)
+		return false;
+
+	pcidev = to_pci_dev(dev);
+
+	bdf = calc_devid(pcidev->bus->number, pcidev->devfn);
+
+	/* Out of our scope? */
+	if (bdf > amd_iommu_last_bdf)
+		return false;
+
+	if (amd_iommu_rlookup_table[bdf] == NULL)
+		return false;
+
 	return true;
 }
 
@@ -2065,22 +2083,7 @@ free_mem:
  */
 static int amd_iommu_dma_supported(struct device *dev, u64 mask)
 {
-	u16 bdf;
-	struct pci_dev *pcidev;
-
-	/* No device or no PCI device */
-	if (!dev || dev->bus != &pci_bus_type)
-		return 0;
-
-	pcidev = to_pci_dev(dev);
-
-	bdf = calc_devid(pcidev->bus->number, pcidev->devfn);
-
-	/* Out of our scope? */
-	if (bdf > amd_iommu_last_bdf)
-		return 0;
-
-	return 1;
+	return check_device(dev);
 }
 
 /*
-- 
1.6.5.3

[PATCH 14/34] x86/amd-iommu: Use check_device in get_device_resources

[Joerg Roedel]

Every call-place of get_device_resources calls check_device
before it. So call it from get_device_resources directly and
simplify the code.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   86 ++++++++++++++-----------------------------
 1 files changed, 28 insertions(+), 58 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index ac27b1d..c5102eb 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1432,35 +1432,24 @@ static struct dma_ops_domain *find_protection_domain(u16 devid)
  * If the device is not yet associated with a domain this is also done
  * in this function.
  */
-static int get_device_resources(struct device *dev,
-				struct amd_iommu **iommu,
-				struct protection_domain **domain,
-				u16 *bdf)
+static bool get_device_resources(struct device *dev,
+				 struct amd_iommu **iommu,
+				 struct protection_domain **domain,
+				 u16 *bdf)
 {
 	struct dma_ops_domain *dma_dom;
 	struct pci_dev *pcidev;
 	u16 _bdf;
 
-	*iommu = NULL;
-	*domain = NULL;
-	*bdf = 0xffff;
-
-	if (dev->bus != &pci_bus_type)
-		return 0;
-
-	pcidev = to_pci_dev(dev);
-	_bdf = calc_devid(pcidev->bus->number, pcidev->devfn);
-
-	/* device not translated by any IOMMU in the system? */
-	if (_bdf > amd_iommu_last_bdf)
-		return 0;
-
-	*bdf = amd_iommu_alias_table[_bdf];
+	if (!check_device(dev))
+		return false;
 
-	*iommu = amd_iommu_rlookup_table[*bdf];
-	if (*iommu == NULL)
-		return 0;
+	pcidev  = to_pci_dev(dev);
+	_bdf    = calc_devid(pcidev->bus->number, pcidev->devfn);
+	*bdf    = amd_iommu_alias_table[_bdf];
+	*iommu  = amd_iommu_rlookup_table[*bdf];
 	*domain = domain_for_device(*bdf);
+
 	if (*domain == NULL) {
 		dma_dom = find_protection_domain(*bdf);
 		if (!dma_dom)
@@ -1474,7 +1463,7 @@ static int get_device_resources(struct device *dev,
 	if (domain_for_device(_bdf) == NULL)
 		attach_device(*iommu, *domain, _bdf);
 
-	return 1;
+	return true;
 }
 
 static void update_device_table(struct protection_domain *domain)
@@ -1797,17 +1786,12 @@ static dma_addr_t map_page(struct device *dev, struct page *page,
 
 	INC_STATS_COUNTER(cnt_map_single);
 
-	if (!check_device(dev))
-		return DMA_ERROR_CODE;
-
-	dma_mask = *dev->dma_mask;
-
-	get_device_resources(dev, &iommu, &domain, &devid);
-
-	if (iommu == NULL || domain == NULL)
+	if (!get_device_resources(dev, &iommu, &domain, &devid))
 		/* device not handled by any AMD IOMMU */
 		return (dma_addr_t)paddr;
 
+	dma_mask = *dev->dma_mask;
+
 	if (!dma_ops_domain(domain))
 		return DMA_ERROR_CODE;
 
@@ -1838,8 +1822,7 @@ static void unmap_page(struct device *dev, dma_addr_t dma_addr, size_t size,
 
 	INC_STATS_COUNTER(cnt_unmap_single);
 
-	if (!check_device(dev) ||
-	    !get_device_resources(dev, &iommu, &domain, &devid))
+	if (!get_device_resources(dev, &iommu, &domain, &devid))
 		/* device not handled by any AMD IOMMU */
 		return;
 
@@ -1893,16 +1876,11 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
 
 	INC_STATS_COUNTER(cnt_map_sg);
 
-	if (!check_device(dev))
-		return 0;
+	if (!get_device_resources(dev, &iommu, &domain, &devid))
+		return map_sg_no_iommu(dev, sglist, nelems, dir);
 
 	dma_mask = *dev->dma_mask;
 
-	get_device_resources(dev, &iommu, &domain, &devid);
-
-	if (!iommu || !domain)
-		return map_sg_no_iommu(dev, sglist, nelems, dir);
-
 	if (!dma_ops_domain(domain))
 		return 0;
 
@@ -1958,8 +1936,7 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
 
 	INC_STATS_COUNTER(cnt_unmap_sg);
 
-	if (!check_device(dev) ||
-	    !get_device_resources(dev, &iommu, &domain, &devid))
+	if (!get_device_resources(dev, &iommu, &domain, &devid))
 		return;
 
 	if (!dma_ops_domain(domain))
@@ -1994,24 +1971,22 @@ static void *alloc_coherent(struct device *dev, size_t size,
 
 	INC_STATS_COUNTER(cnt_alloc_coherent);
 
-	if (!check_device(dev))
-		return NULL;
+	if (!get_device_resources(dev, &iommu, &domain, &devid)) {
+		virt_addr = (void *)__get_free_pages(flag, get_order(size));
+		*dma_addr = __pa(virt_addr);
+		return virt_addr;
+	}
 
-	if (!get_device_resources(dev, &iommu, &domain, &devid))
-		flag &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
+	dma_mask  = dev->coherent_dma_mask;
+	flag     &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
+	flag     |= __GFP_ZERO;
 
-	flag |= __GFP_ZERO;
 	virt_addr = (void *)__get_free_pages(flag, get_order(size));
 	if (!virt_addr)
 		return NULL;
 
 	paddr = virt_to_phys(virt_addr);
 
-	if (!iommu || !domain) {
-		*dma_addr = (dma_addr_t)paddr;
-		return virt_addr;
-	}
-
 	if (!dma_ops_domain(domain))
 		goto out_free;
 
@@ -2054,12 +2029,7 @@ static void free_coherent(struct device *dev, size_t size,
 
 	INC_STATS_COUNTER(cnt_free_coherent);
 
-	if (!check_device(dev))
-		return;
-
-	get_device_resources(dev, &iommu, &domain, &devid);
-
-	if (!iommu || !domain)
+	if (!get_device_resources(dev, &iommu, &domain, &devid))
 		goto free_mem;
 
 	if (!dma_ops_domain(domain))
-- 
1.6.5.3

[PATCH 15/34] x86/amd-iommu: Remove iommu parameter from dma_ops_domain_(un)map

[Joerg Roedel]

The parameter is unused in these function so remove it from
the parameter list.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   12 +++++-------
 1 files changed, 5 insertions(+), 7 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index c5102eb..da3f9d8 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1585,8 +1585,7 @@ static u64* dma_ops_get_pte(struct dma_ops_domain *dom,
  * This is the generic map function. It maps one 4kb page at paddr to
  * the given address in the DMA address space for the domain.
  */
-static dma_addr_t dma_ops_domain_map(struct amd_iommu *iommu,
-				     struct dma_ops_domain *dom,
+static dma_addr_t dma_ops_domain_map(struct dma_ops_domain *dom,
 				     unsigned long address,
 				     phys_addr_t paddr,
 				     int direction)
@@ -1620,8 +1619,7 @@ static dma_addr_t dma_ops_domain_map(struct amd_iommu *iommu,
 /*
  * The generic unmapping function for on page in the DMA address space.
  */
-static void dma_ops_domain_unmap(struct amd_iommu *iommu,
-				 struct dma_ops_domain *dom,
+static void dma_ops_domain_unmap(struct dma_ops_domain *dom,
 				 unsigned long address)
 {
 	struct aperture_range *aperture;
@@ -1700,7 +1698,7 @@ retry:
 
 	start = address;
 	for (i = 0; i < pages; ++i) {
-		ret = dma_ops_domain_map(iommu, dma_dom, start, paddr, dir);
+		ret = dma_ops_domain_map(dma_dom, start, paddr, dir);
 		if (ret == DMA_ERROR_CODE)
 			goto out_unmap;
 
@@ -1724,7 +1722,7 @@ out_unmap:
 
 	for (--i; i >= 0; --i) {
 		start -= PAGE_SIZE;
-		dma_ops_domain_unmap(iommu, dma_dom, start);
+		dma_ops_domain_unmap(dma_dom, start);
 	}
 
 	dma_ops_free_addresses(dma_dom, address, pages);
@@ -1754,7 +1752,7 @@ static void __unmap_single(struct amd_iommu *iommu,
 	start = dma_addr;
 
 	for (i = 0; i < pages; ++i) {
-		dma_ops_domain_unmap(iommu, dma_dom, start);
+		dma_ops_domain_unmap(dma_dom, start);
 		start += PAGE_SIZE;
 	}
 
-- 
1.6.5.3

[PATCH 16/34] x86/amd-iommu: Make alloc_new_range aware of multiple IOMMUs

[Joerg Roedel]

Since the assumption that an dma_ops domain is only bound to
one IOMMU was given up we need to make alloc_new_range aware
of it.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   27 +++++++++++++++------------
 1 files changed, 15 insertions(+), 12 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index da3f9d8..687f617 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -788,11 +788,11 @@ static u64 *fetch_pte(struct protection_domain *domain,
  * aperture in case of dma_ops domain allocation or address allocation
  * failure.
  */
-static int alloc_new_range(struct amd_iommu *iommu,
-			   struct dma_ops_domain *dma_dom,
+static int alloc_new_range(struct dma_ops_domain *dma_dom,
 			   bool populate, gfp_t gfp)
 {
 	int index = dma_dom->aperture_size >> APERTURE_RANGE_SHIFT;
+	struct amd_iommu *iommu;
 	int i;
 
 #ifdef CONFIG_IOMMU_STRESS
@@ -832,14 +832,17 @@ static int alloc_new_range(struct amd_iommu *iommu,
 	dma_dom->aperture_size += APERTURE_RANGE_SIZE;
 
 	/* Intialize the exclusion range if necessary */
-	if (iommu->exclusion_start &&
-	    iommu->exclusion_start >= dma_dom->aperture[index]->offset &&
-	    iommu->exclusion_start < dma_dom->aperture_size) {
-		unsigned long startpage = iommu->exclusion_start >> PAGE_SHIFT;
-		int pages = iommu_num_pages(iommu->exclusion_start,
-					    iommu->exclusion_length,
-					    PAGE_SIZE);
-		dma_ops_reserve_addresses(dma_dom, startpage, pages);
+	for_each_iommu(iommu) {
+		if (iommu->exclusion_start &&
+		    iommu->exclusion_start >= dma_dom->aperture[index]->offset
+		    && iommu->exclusion_start < dma_dom->aperture_size) {
+			unsigned long startpage;
+			int pages = iommu_num_pages(iommu->exclusion_start,
+						    iommu->exclusion_length,
+						    PAGE_SIZE);
+			startpage = iommu->exclusion_start >> PAGE_SHIFT;
+			dma_ops_reserve_addresses(dma_dom, startpage, pages);
+		}
 	}
 
 	/*
@@ -1143,7 +1146,7 @@ static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu)
 
 	add_domain_to_list(&dma_dom->domain);
 
-	if (alloc_new_range(iommu, dma_dom, true, GFP_KERNEL))
+	if (alloc_new_range(dma_dom, true, GFP_KERNEL))
 		goto free_dma_dom;
 
 	/*
@@ -1686,7 +1689,7 @@ retry:
 		 */
 		dma_dom->next_address = dma_dom->aperture_size;
 
-		if (alloc_new_range(iommu, dma_dom, false, GFP_ATOMIC))
+		if (alloc_new_range(dma_dom, false, GFP_ATOMIC))
 			goto out;
 
 		/*
-- 
1.6.5.3

[PATCH 17/34] x86/amd-iommu: Remove iommu parameter from __(un)map_single

[Joerg Roedel]

With the prior changes this parameter is not longer
required. This patch removes it from the function and all
callers.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   18 ++++++++----------
 1 files changed, 8 insertions(+), 10 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 687f617..c04dcb7 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1653,7 +1653,6 @@ static void dma_ops_domain_unmap(struct dma_ops_domain *dom,
  * Must be called with the domain lock held.
  */
 static dma_addr_t __map_single(struct device *dev,
-			       struct amd_iommu *iommu,
 			       struct dma_ops_domain *dma_dom,
 			       phys_addr_t paddr,
 			       size_t size,
@@ -1737,8 +1736,7 @@ out_unmap:
  * Does the reverse of the __map_single function. Must be called with
  * the domain lock held too
  */
-static void __unmap_single(struct amd_iommu *iommu,
-			   struct dma_ops_domain *dma_dom,
+static void __unmap_single(struct dma_ops_domain *dma_dom,
 			   dma_addr_t dma_addr,
 			   size_t size,
 			   int dir)
@@ -1797,7 +1795,7 @@ static dma_addr_t map_page(struct device *dev, struct page *page,
 		return DMA_ERROR_CODE;
 
 	spin_lock_irqsave(&domain->lock, flags);
-	addr = __map_single(dev, iommu, domain->priv, paddr, size, dir, false,
+	addr = __map_single(dev, domain->priv, paddr, size, dir, false,
 			    dma_mask);
 	if (addr == DMA_ERROR_CODE)
 		goto out;
@@ -1832,7 +1830,7 @@ static void unmap_page(struct device *dev, dma_addr_t dma_addr, size_t size,
 
 	spin_lock_irqsave(&domain->lock, flags);
 
-	__unmap_single(iommu, domain->priv, dma_addr, size, dir);
+	__unmap_single(domain->priv, dma_addr, size, dir);
 
 	iommu_flush_complete(domain);
 
@@ -1890,7 +1888,7 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
 	for_each_sg(sglist, s, nelems, i) {
 		paddr = sg_phys(s);
 
-		s->dma_address = __map_single(dev, iommu, domain->priv,
+		s->dma_address = __map_single(dev, domain->priv,
 					      paddr, s->length, dir, false,
 					      dma_mask);
 
@@ -1910,7 +1908,7 @@ out:
 unmap:
 	for_each_sg(sglist, s, mapped_elems, i) {
 		if (s->dma_address)
-			__unmap_single(iommu, domain->priv, s->dma_address,
+			__unmap_single(domain->priv, s->dma_address,
 				       s->dma_length, dir);
 		s->dma_address = s->dma_length = 0;
 	}
@@ -1946,7 +1944,7 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
 	spin_lock_irqsave(&domain->lock, flags);
 
 	for_each_sg(sglist, s, nelems, i) {
-		__unmap_single(iommu, domain->priv, s->dma_address,
+		__unmap_single(domain->priv, s->dma_address,
 			       s->dma_length, dir);
 		s->dma_address = s->dma_length = 0;
 	}
@@ -1996,7 +1994,7 @@ static void *alloc_coherent(struct device *dev, size_t size,
 
 	spin_lock_irqsave(&domain->lock, flags);
 
-	*dma_addr = __map_single(dev, iommu, domain->priv, paddr,
+	*dma_addr = __map_single(dev, domain->priv, paddr,
 				 size, DMA_BIDIRECTIONAL, true, dma_mask);
 
 	if (*dma_addr == DMA_ERROR_CODE) {
@@ -2038,7 +2036,7 @@ static void free_coherent(struct device *dev, size_t size,
 
 	spin_lock_irqsave(&domain->lock, flags);
 
-	__unmap_single(iommu, domain->priv, dma_addr, size, DMA_BIDIRECTIONAL);
+	__unmap_single(domain->priv, dma_addr, size, DMA_BIDIRECTIONAL);
 
 	iommu_flush_complete(domain);
 
-- 
1.6.5.3

[PATCH 18/34] x86/amd-iommu: Remove iommu specific handling from dma_ops path

[Joerg Roedel]

This patch finishes the removal of all iommu specific
handling code in the dma_ops path.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   28 +++++++++++-----------------
 1 files changed, 11 insertions(+), 17 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index c04dcb7..2cd5800 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1436,11 +1436,11 @@ static struct dma_ops_domain *find_protection_domain(u16 devid)
  * in this function.
  */
 static bool get_device_resources(struct device *dev,
-				 struct amd_iommu **iommu,
 				 struct protection_domain **domain,
 				 u16 *bdf)
 {
 	struct dma_ops_domain *dma_dom;
+	struct amd_iommu *iommu;
 	struct pci_dev *pcidev;
 	u16 _bdf;
 
@@ -1450,21 +1450,21 @@ static bool get_device_resources(struct device *dev,
 	pcidev  = to_pci_dev(dev);
 	_bdf    = calc_devid(pcidev->bus->number, pcidev->devfn);
 	*bdf    = amd_iommu_alias_table[_bdf];
-	*iommu  = amd_iommu_rlookup_table[*bdf];
+	iommu   = amd_iommu_rlookup_table[*bdf];
 	*domain = domain_for_device(*bdf);
 
 	if (*domain == NULL) {
 		dma_dom = find_protection_domain(*bdf);
 		if (!dma_dom)
-			dma_dom = (*iommu)->default_dom;
+			dma_dom = iommu->default_dom;
 		*domain = &dma_dom->domain;
-		attach_device(*iommu, *domain, *bdf);
+		attach_device(iommu, *domain, *bdf);
 		DUMP_printk("Using protection domain %d for device %s\n",
 			    (*domain)->id, dev_name(dev));
 	}
 
 	if (domain_for_device(_bdf) == NULL)
-		attach_device(*iommu, *domain, _bdf);
+		attach_device(iommu, *domain, _bdf);
 
 	return true;
 }
@@ -1776,7 +1776,6 @@ static dma_addr_t map_page(struct device *dev, struct page *page,
 			   struct dma_attrs *attrs)
 {
 	unsigned long flags;
-	struct amd_iommu *iommu;
 	struct protection_domain *domain;
 	u16 devid;
 	dma_addr_t addr;
@@ -1785,7 +1784,7 @@ static dma_addr_t map_page(struct device *dev, struct page *page,
 
 	INC_STATS_COUNTER(cnt_map_single);
 
-	if (!get_device_resources(dev, &iommu, &domain, &devid))
+	if (!get_device_resources(dev, &domain, &devid))
 		/* device not handled by any AMD IOMMU */
 		return (dma_addr_t)paddr;
 
@@ -1815,13 +1814,12 @@ static void unmap_page(struct device *dev, dma_addr_t dma_addr, size_t size,
 		       enum dma_data_direction dir, struct dma_attrs *attrs)
 {
 	unsigned long flags;
-	struct amd_iommu *iommu;
 	struct protection_domain *domain;
 	u16 devid;
 
 	INC_STATS_COUNTER(cnt_unmap_single);
 
-	if (!get_device_resources(dev, &iommu, &domain, &devid))
+	if (!get_device_resources(dev, &domain, &devid))
 		/* device not handled by any AMD IOMMU */
 		return;
 
@@ -1864,7 +1862,6 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
 		  struct dma_attrs *attrs)
 {
 	unsigned long flags;
-	struct amd_iommu *iommu;
 	struct protection_domain *domain;
 	u16 devid;
 	int i;
@@ -1875,7 +1872,7 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
 
 	INC_STATS_COUNTER(cnt_map_sg);
 
-	if (!get_device_resources(dev, &iommu, &domain, &devid))
+	if (!get_device_resources(dev, &domain, &devid))
 		return map_sg_no_iommu(dev, sglist, nelems, dir);
 
 	dma_mask = *dev->dma_mask;
@@ -1927,7 +1924,6 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
 		     struct dma_attrs *attrs)
 {
 	unsigned long flags;
-	struct amd_iommu *iommu;
 	struct protection_domain *domain;
 	struct scatterlist *s;
 	u16 devid;
@@ -1935,7 +1931,7 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
 
 	INC_STATS_COUNTER(cnt_unmap_sg);
 
-	if (!get_device_resources(dev, &iommu, &domain, &devid))
+	if (!get_device_resources(dev, &domain, &devid))
 		return;
 
 	if (!dma_ops_domain(domain))
@@ -1962,7 +1958,6 @@ static void *alloc_coherent(struct device *dev, size_t size,
 {
 	unsigned long flags;
 	void *virt_addr;
-	struct amd_iommu *iommu;
 	struct protection_domain *domain;
 	u16 devid;
 	phys_addr_t paddr;
@@ -1970,7 +1965,7 @@ static void *alloc_coherent(struct device *dev, size_t size,
 
 	INC_STATS_COUNTER(cnt_alloc_coherent);
 
-	if (!get_device_resources(dev, &iommu, &domain, &devid)) {
+	if (!get_device_resources(dev, &domain, &devid)) {
 		virt_addr = (void *)__get_free_pages(flag, get_order(size));
 		*dma_addr = __pa(virt_addr);
 		return virt_addr;
@@ -2022,13 +2017,12 @@ static void free_coherent(struct device *dev, size_t size,
 			  void *virt_addr, dma_addr_t dma_addr)
 {
 	unsigned long flags;
-	struct amd_iommu *iommu;
 	struct protection_domain *domain;
 	u16 devid;
 
 	INC_STATS_COUNTER(cnt_free_coherent);
 
-	if (!get_device_resources(dev, &iommu, &domain, &devid))
+	if (!get_device_resources(dev, &domain, &devid))
 		goto free_mem;
 
 	if (!dma_ops_domain(domain))
-- 
1.6.5.3

[PATCH 19/34] x86/amd-iommu: Let domain_for_device handle aliases

[Joerg Roedel]

If there is no domain associated to a device yet and the
device has an alias device which already has a domain, the
original device needs to have the same domain as the alias
device.
This patch changes domain_for_device to handle this
situation and directly assigns the alias device domain to
the device in this situation.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |  227 +++++++++++++++++++++++++-----------------
 1 files changed, 135 insertions(+), 92 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 2cd5800..75470ff 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -71,6 +71,19 @@ static u64 *fetch_pte(struct protection_domain *domain,
 		      unsigned long address, int map_size);
 static void update_domain(struct protection_domain *domain);
 
+/****************************************************************************
+ *
+ * Helper functions
+ *
+ ****************************************************************************/
+
+static inline u16 get_device_id(struct device *dev)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
+
+	return calc_devid(pdev->bus->number, pdev->devfn);
+}
+
 #ifdef CONFIG_AMD_IOMMU_STATS
 
 /*
@@ -1174,26 +1187,13 @@ static bool dma_ops_domain(struct protection_domain *domain)
 	return domain->flags & PD_DMA_OPS_MASK;
 }
 
-/*
- * Find out the protection domain structure for a given PCI device. This
- * will give us the pointer to the page table root for example.
- */
-static struct protection_domain *domain_for_device(u16 devid)
-{
-	struct protection_domain *dom;
-	unsigned long flags;
-
-	read_lock_irqsave(&amd_iommu_devtable_lock, flags);
-	dom = amd_iommu_pd_table[devid];
-	read_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
-
-	return dom;
-}
-
 static void set_dte_entry(u16 devid, struct protection_domain *domain)
 {
+	struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
 	u64 pte_root = virt_to_phys(domain->pt_root);
 
+	BUG_ON(amd_iommu_pd_table[devid] != NULL);
+
 	pte_root |= (domain->mode & DEV_ENTRY_MODE_MASK)
 		    << DEV_ENTRY_MODE_SHIFT;
 	pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | IOMMU_PTE_TV;
@@ -1203,42 +1203,87 @@ static void set_dte_entry(u16 devid, struct protection_domain *domain)
 	amd_iommu_dev_table[devid].data[0] = lower_32_bits(pte_root);
 
 	amd_iommu_pd_table[devid] = domain;
+
+	/* Do reference counting */
+	domain->dev_iommu[iommu->index] += 1;
+	domain->dev_cnt                 += 1;
+
+	/* Flush the changes DTE entry */
+	iommu_queue_inv_dev_entry(iommu, devid);
+}
+
+static void clear_dte_entry(u16 devid)
+{
+	struct protection_domain *domain = amd_iommu_pd_table[devid];
+	struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
+
+	BUG_ON(domain == NULL);
+
+	/* remove domain from the lookup table */
+	amd_iommu_pd_table[devid] = NULL;
+
+	/* remove entry from the device table seen by the hardware */
+	amd_iommu_dev_table[devid].data[0] = IOMMU_PTE_P | IOMMU_PTE_TV;
+	amd_iommu_dev_table[devid].data[1] = 0;
+	amd_iommu_dev_table[devid].data[2] = 0;
+
+	amd_iommu_apply_erratum_63(devid);
+
+	/* decrease reference counters */
+	domain->dev_iommu[iommu->index] -= 1;
+	domain->dev_cnt                 -= 1;
+
+	iommu_queue_inv_dev_entry(iommu, devid);
 }
 
 /*
  * If a device is not yet associated with a domain, this function does
  * assigns it visible for the hardware
  */
-static void __attach_device(struct amd_iommu *iommu,
-			    struct protection_domain *domain,
-			    u16 devid)
+static int __attach_device(struct device *dev,
+			   struct protection_domain *domain)
 {
+	u16 devid = get_device_id(dev);
+	u16 alias = amd_iommu_alias_table[devid];
+
 	/* lock domain */
 	spin_lock(&domain->lock);
 
-	/* update DTE entry */
-	set_dte_entry(devid, domain);
+	/* Some sanity checks */
+	if (amd_iommu_pd_table[alias] != NULL &&
+	    amd_iommu_pd_table[alias] != domain)
+		return -EBUSY;
 
-	/* Do reference counting */
-	domain->dev_iommu[iommu->index] += 1;
-	domain->dev_cnt                 += 1;
+	if (amd_iommu_pd_table[devid] != NULL &&
+	    amd_iommu_pd_table[devid] != domain)
+		return -EBUSY;
+
+	/* Do real assignment */
+	if (alias != devid &&
+	    amd_iommu_pd_table[alias] == NULL)
+		set_dte_entry(alias, domain);
+
+	if (amd_iommu_pd_table[devid] == NULL)
+		set_dte_entry(devid, domain);
 
 	/* ready */
 	spin_unlock(&domain->lock);
+
+	return 0;
 }
 
 /*
  * If a device is not yet associated with a domain, this function does
  * assigns it visible for the hardware
  */
-static void attach_device(struct amd_iommu *iommu,
-			  struct protection_domain *domain,
-			  u16 devid)
+static int attach_device(struct device *dev,
+			 struct protection_domain *domain)
 {
 	unsigned long flags;
+	int ret;
 
 	write_lock_irqsave(&amd_iommu_devtable_lock, flags);
-	__attach_device(iommu, domain, devid);
+	ret = __attach_device(dev, domain);
 	write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
 
 	/*
@@ -1246,62 +1291,70 @@ static void attach_device(struct amd_iommu *iommu,
 	 * left the caches in the IOMMU dirty. So we have to flush
 	 * here to evict all dirty stuff.
 	 */
-	iommu_queue_inv_dev_entry(iommu, devid);
 	iommu_flush_tlb_pde(domain);
+
+	return ret;
 }
 
 /*
  * Removes a device from a protection domain (unlocked)
  */
-static void __detach_device(struct protection_domain *domain, u16 devid)
+static void __detach_device(struct device *dev)
 {
+	u16 devid = get_device_id(dev);
 	struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
 
 	BUG_ON(!iommu);
 
-	/* lock domain */
-	spin_lock(&domain->lock);
-
-	/* remove domain from the lookup table */
-	amd_iommu_pd_table[devid] = NULL;
-
-	/* remove entry from the device table seen by the hardware */
-	amd_iommu_dev_table[devid].data[0] = IOMMU_PTE_P | IOMMU_PTE_TV;
-	amd_iommu_dev_table[devid].data[1] = 0;
-	amd_iommu_dev_table[devid].data[2] = 0;
-
-	amd_iommu_apply_erratum_63(devid);
-
-	/* decrease reference counters */
-	domain->dev_iommu[iommu->index] -= 1;
-	domain->dev_cnt                 -= 1;
-
-	/* ready */
-	spin_unlock(&domain->lock);
+	clear_dte_entry(devid);
 
 	/*
 	 * If we run in passthrough mode the device must be assigned to the
 	 * passthrough domain if it is detached from any other domain
 	 */
-	if (iommu_pass_through) {
-		struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
-		__attach_device(iommu, pt_domain, devid);
-	}
+	if (iommu_pass_through)
+		__attach_device(dev, pt_domain);
 }
 
 /*
  * Removes a device from a protection domain (with devtable_lock held)
  */
-static void detach_device(struct protection_domain *domain, u16 devid)
+static void detach_device(struct device *dev)
 {
 	unsigned long flags;
 
 	/* lock device table */
 	write_lock_irqsave(&amd_iommu_devtable_lock, flags);
-	__detach_device(domain, devid);
+	__detach_device(dev);
 	write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
 }
 
+/*
+ * Find out the protection domain structure for a given PCI device. This
+ * will give us the pointer to the page table root for example.
+ */
+static struct protection_domain *domain_for_device(struct device *dev)
+{
+	struct protection_domain *dom;
+	unsigned long flags;
+	u16 devid, alias;
+
+	devid = get_device_id(dev);
+	alias = amd_iommu_alias_table[devid];
+
+	read_lock_irqsave(&amd_iommu_devtable_lock, flags);
+	dom = amd_iommu_pd_table[devid];
+	if (dom == NULL &&
+	    amd_iommu_pd_table[alias] != NULL) {
+		__attach_device(dev, amd_iommu_pd_table[alias]);
+		dom = amd_iommu_pd_table[devid];
+	}
+
+	read_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
+
+	return dom;
+}
+
 static int device_change_notifier(struct notifier_block *nb,
 				  unsigned long action, void *data)
 {
@@ -1322,7 +1375,7 @@ static int device_change_notifier(struct notifier_block *nb,
 	if (iommu == NULL)
 		goto out;
 
-	domain = domain_for_device(devid);
+	domain = domain_for_device(dev);
 
 	if (domain && !dma_ops_domain(domain))
 		WARN_ONCE(1, "AMD IOMMU WARNING: device %s already bound "
@@ -1334,7 +1387,7 @@ static int device_change_notifier(struct notifier_block *nb,
 			goto out;
 		if (iommu_pass_through)
 			break;
-		detach_device(domain, devid);
+		detach_device(dev);
 		break;
 	case BUS_NOTIFY_ADD_DEVICE:
 		/* allocate a protection domain if a device is added */
@@ -1441,30 +1494,25 @@ static bool get_device_resources(struct device *dev,
 {
 	struct dma_ops_domain *dma_dom;
 	struct amd_iommu *iommu;
-	struct pci_dev *pcidev;
-	u16 _bdf;
 
 	if (!check_device(dev))
 		return false;
 
-	pcidev  = to_pci_dev(dev);
-	_bdf    = calc_devid(pcidev->bus->number, pcidev->devfn);
-	*bdf    = amd_iommu_alias_table[_bdf];
+	*bdf    = get_device_id(dev);
+	*domain = domain_for_device(dev);
 	iommu   = amd_iommu_rlookup_table[*bdf];
-	*domain = domain_for_device(*bdf);
 
-	if (*domain == NULL) {
-		dma_dom = find_protection_domain(*bdf);
-		if (!dma_dom)
-			dma_dom = iommu->default_dom;
-		*domain = &dma_dom->domain;
-		attach_device(iommu, *domain, *bdf);
-		DUMP_printk("Using protection domain %d for device %s\n",
-			    (*domain)->id, dev_name(dev));
-	}
+	if (*domain != NULL)
+		return true;
 
-	if (domain_for_device(_bdf) == NULL)
-		attach_device(iommu, *domain, _bdf);
+	/* Device not bount yet - bind it */
+	dma_dom = find_protection_domain(*bdf);
+	if (!dma_dom)
+		dma_dom = iommu->default_dom;
+	*domain = &dma_dom->domain;
+	attach_device(dev, *domain);
+	DUMP_printk("Using protection domain %d for device %s\n",
+		    (*domain)->id, dev_name(dev));
 
 	return true;
 }
@@ -2068,7 +2116,7 @@ static void prealloc_protection_domains(void)
 		if (devid > amd_iommu_last_bdf)
 			continue;
 		devid = amd_iommu_alias_table[devid];
-		if (domain_for_device(devid))
+		if (domain_for_device(&dev->dev))
 			continue;
 		iommu = amd_iommu_rlookup_table[devid];
 		if (!iommu)
@@ -2079,9 +2127,7 @@ static void prealloc_protection_domains(void)
 		init_unity_mappings_for_device(dma_dom, devid);
 		dma_dom->target_dev = devid;
 
-		attach_device(iommu, &dma_dom->domain, devid);
-		if (__devid != devid)
-			attach_device(iommu, &dma_dom->domain, __devid);
+		attach_device(&dev->dev, &dma_dom->domain);
 
 		list_add_tail(&dma_dom->list, &iommu_pd_list);
 	}
@@ -2174,7 +2220,7 @@ static void cleanup_domain(struct protection_domain *domain)
 
 	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
 		if (amd_iommu_pd_table[devid] == domain)
-			__detach_device(domain, devid);
+			clear_dte_entry(devid);
 
 	write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
 }
@@ -2262,7 +2308,6 @@ static void amd_iommu_domain_destroy(struct iommu_domain *dom)
 static void amd_iommu_detach_device(struct iommu_domain *dom,
 				    struct device *dev)
 {
-	struct protection_domain *domain = dom->priv;
 	struct amd_iommu *iommu;
 	struct pci_dev *pdev;
 	u16 devid;
@@ -2275,7 +2320,7 @@ static void amd_iommu_detach_device(struct iommu_domain *dom,
 	devid = calc_devid(pdev->bus->number, pdev->devfn);
 
 	if (devid > 0)
-		detach_device(domain, devid);
+		detach_device(dev);
 
 	iommu = amd_iommu_rlookup_table[devid];
 	if (!iommu)
@@ -2292,6 +2337,7 @@ static int amd_iommu_attach_device(struct iommu_domain *dom,
 	struct protection_domain *old_domain;
 	struct amd_iommu *iommu;
 	struct pci_dev *pdev;
+	int ret;
 	u16 devid;
 
 	if (dev->bus != &pci_bus_type)
@@ -2309,15 +2355,15 @@ static int amd_iommu_attach_device(struct iommu_domain *dom,
 	if (!iommu)
 		return -EINVAL;
 
-	old_domain = domain_for_device(devid);
+	old_domain = amd_iommu_pd_table[devid];
 	if (old_domain)
-		detach_device(old_domain, devid);
+		detach_device(dev);
 
-	attach_device(iommu, domain, devid);
+	ret = attach_device(dev, domain);
 
 	iommu_completion_wait(iommu);
 
-	return 0;
+	return ret;
 }
 
 static int amd_iommu_map_range(struct iommu_domain *dom,
@@ -2414,8 +2460,9 @@ static struct iommu_ops amd_iommu_ops = {
 
 int __init amd_iommu_init_passthrough(void)
 {
+	struct amd_iommu *iommu;
 	struct pci_dev *dev = NULL;
-	u16 devid, devid2;
+	u16 devid;
 
 	/* allocate passthroug domain */
 	pt_domain = protection_domain_alloc();
@@ -2425,20 +2472,16 @@ int __init amd_iommu_init_passthrough(void)
 	pt_domain->mode |= PAGE_MODE_NONE;
 
 	while ((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) {
-		struct amd_iommu *iommu;
 
 		devid = calc_devid(dev->bus->number, dev->devfn);
 		if (devid > amd_iommu_last_bdf)
 			continue;
 
-		devid2 = amd_iommu_alias_table[devid];
-
-		iommu = amd_iommu_rlookup_table[devid2];
+		iommu = amd_iommu_rlookup_table[devid];
 		if (!iommu)
 			continue;
 
-		__attach_device(iommu, pt_domain, devid);
-		__attach_device(iommu, pt_domain, devid2);
+		attach_device(&dev->dev, pt_domain);
 	}
 
 	pr_info("AMD-Vi: Initialized for Passthrough Mode\n");
-- 
1.6.5.3

[PATCH 20/34] x86/amd-iommu: Simplify get_device_resources()

[Joerg Roedel]

With the previous changes the get_device_resources function
can be simplified even more. The only important information
for the callers is the protection domain.
This patch renames the function to get_domain() and let it
only return the protection domain for a device.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   86 ++++++++++++++++++-------------------------
 1 files changed, 36 insertions(+), 50 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 75470ff..e5bbe9a 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1463,6 +1463,7 @@ static struct dma_ops_domain *find_protection_domain(u16 devid)
 {
 	struct dma_ops_domain *entry, *ret = NULL;
 	unsigned long flags;
+	u16 alias = amd_iommu_alias_table[devid];
 
 	if (list_empty(&iommu_pd_list))
 		return NULL;
@@ -1470,7 +1471,8 @@ static struct dma_ops_domain *find_protection_domain(u16 devid)
 	spin_lock_irqsave(&iommu_pd_list_lock, flags);
 
 	list_for_each_entry(entry, &iommu_pd_list, list) {
-		if (entry->target_dev == devid) {
+		if (entry->target_dev == devid ||
+		    entry->target_dev == alias) {
 			ret = entry;
 			break;
 		}
@@ -1488,33 +1490,31 @@ static struct dma_ops_domain *find_protection_domain(u16 devid)
  * If the device is not yet associated with a domain this is also done
  * in this function.
  */
-static bool get_device_resources(struct device *dev,
-				 struct protection_domain **domain,
-				 u16 *bdf)
+static struct protection_domain *get_domain(struct device *dev)
 {
+	struct protection_domain *domain;
 	struct dma_ops_domain *dma_dom;
-	struct amd_iommu *iommu;
+	u16 devid = get_device_id(dev);
 
 	if (!check_device(dev))
-		return false;
+		return ERR_PTR(-EINVAL);
 
-	*bdf    = get_device_id(dev);
-	*domain = domain_for_device(dev);
-	iommu   = amd_iommu_rlookup_table[*bdf];
+	domain = domain_for_device(dev);
+	if (domain != NULL && !dma_ops_domain(domain))
+		return ERR_PTR(-EBUSY);
 
-	if (*domain != NULL)
-		return true;
+	if (domain != NULL)
+		return domain;
 
 	/* Device not bount yet - bind it */
-	dma_dom = find_protection_domain(*bdf);
+	dma_dom = find_protection_domain(devid);
 	if (!dma_dom)
-		dma_dom = iommu->default_dom;
-	*domain = &dma_dom->domain;
-	attach_device(dev, *domain);
+		dma_dom = amd_iommu_rlookup_table[devid]->default_dom;
+	attach_device(dev, &dma_dom->domain);
 	DUMP_printk("Using protection domain %d for device %s\n",
-		    (*domain)->id, dev_name(dev));
+		    dma_dom->domain.id, dev_name(dev));
 
-	return true;
+	return &dma_dom->domain;
 }
 
 static void update_device_table(struct protection_domain *domain)
@@ -1825,23 +1825,22 @@ static dma_addr_t map_page(struct device *dev, struct page *page,
 {
 	unsigned long flags;
 	struct protection_domain *domain;
-	u16 devid;
 	dma_addr_t addr;
 	u64 dma_mask;
 	phys_addr_t paddr = page_to_phys(page) + offset;
 
 	INC_STATS_COUNTER(cnt_map_single);
 
-	if (!get_device_resources(dev, &domain, &devid))
-		/* device not handled by any AMD IOMMU */
+	domain = get_domain(dev);
+	if (PTR_ERR(domain) == -EINVAL)
 		return (dma_addr_t)paddr;
+	else if (IS_ERR(domain))
+		return DMA_ERROR_CODE;
 
 	dma_mask = *dev->dma_mask;
 
-	if (!dma_ops_domain(domain))
-		return DMA_ERROR_CODE;
-
 	spin_lock_irqsave(&domain->lock, flags);
+
 	addr = __map_single(dev, domain->priv, paddr, size, dir, false,
 			    dma_mask);
 	if (addr == DMA_ERROR_CODE)
@@ -1863,15 +1862,11 @@ static void unmap_page(struct device *dev, dma_addr_t dma_addr, size_t size,
 {
 	unsigned long flags;
 	struct protection_domain *domain;
-	u16 devid;
 
 	INC_STATS_COUNTER(cnt_unmap_single);
 
-	if (!get_device_resources(dev, &domain, &devid))
-		/* device not handled by any AMD IOMMU */
-		return;
-
-	if (!dma_ops_domain(domain))
+	domain = get_domain(dev);
+	if (IS_ERR(domain))
 		return;
 
 	spin_lock_irqsave(&domain->lock, flags);
@@ -1911,7 +1906,6 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
 {
 	unsigned long flags;
 	struct protection_domain *domain;
-	u16 devid;
 	int i;
 	struct scatterlist *s;
 	phys_addr_t paddr;
@@ -1920,14 +1914,14 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
 
 	INC_STATS_COUNTER(cnt_map_sg);
 
-	if (!get_device_resources(dev, &domain, &devid))
+	domain = get_domain(dev);
+	if (PTR_ERR(domain) == -EINVAL)
 		return map_sg_no_iommu(dev, sglist, nelems, dir);
+	else if (IS_ERR(domain))
+		return 0;
 
 	dma_mask = *dev->dma_mask;
 
-	if (!dma_ops_domain(domain))
-		return 0;
-
 	spin_lock_irqsave(&domain->lock, flags);
 
 	for_each_sg(sglist, s, nelems, i) {
@@ -1974,15 +1968,12 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
 	unsigned long flags;
 	struct protection_domain *domain;
 	struct scatterlist *s;
-	u16 devid;
 	int i;
 
 	INC_STATS_COUNTER(cnt_unmap_sg);
 
-	if (!get_device_resources(dev, &domain, &devid))
-		return;
-
-	if (!dma_ops_domain(domain))
+	domain = get_domain(dev);
+	if (IS_ERR(domain))
 		return;
 
 	spin_lock_irqsave(&domain->lock, flags);
@@ -2007,17 +1998,18 @@ static void *alloc_coherent(struct device *dev, size_t size,
 	unsigned long flags;
 	void *virt_addr;
 	struct protection_domain *domain;
-	u16 devid;
 	phys_addr_t paddr;
 	u64 dma_mask = dev->coherent_dma_mask;
 
 	INC_STATS_COUNTER(cnt_alloc_coherent);
 
-	if (!get_device_resources(dev, &domain, &devid)) {
+	domain = get_domain(dev);
+	if (PTR_ERR(domain) == -EINVAL) {
 		virt_addr = (void *)__get_free_pages(flag, get_order(size));
 		*dma_addr = __pa(virt_addr);
 		return virt_addr;
-	}
+	} else if (IS_ERR(domain))
+		return NULL;
 
 	dma_mask  = dev->coherent_dma_mask;
 	flag     &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
@@ -2029,9 +2021,6 @@ static void *alloc_coherent(struct device *dev, size_t size,
 
 	paddr = virt_to_phys(virt_addr);
 
-	if (!dma_ops_domain(domain))
-		goto out_free;
-
 	if (!dma_mask)
 		dma_mask = *dev->dma_mask;
 
@@ -2066,14 +2055,11 @@ static void free_coherent(struct device *dev, size_t size,
 {
 	unsigned long flags;
 	struct protection_domain *domain;
-	u16 devid;
 
 	INC_STATS_COUNTER(cnt_free_coherent);
 
-	if (!get_device_resources(dev, &domain, &devid))
-		goto free_mem;
-
-	if (!dma_ops_domain(domain))
+	domain = get_domain(dev);
+	if (IS_ERR(domain))
 		goto free_mem;
 
 	spin_lock_irqsave(&domain->lock, flags);
-- 
1.6.5.3

[PATCH 21/34] x86/amd-iommu: Move find_protection_domain to helper functions

[Joerg Roedel]

This is a helper function and when its placed in the helper
function section we can remove its forward declaration.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   57 +++++++++++++++++++++----------------------
 1 files changed, 28 insertions(+), 29 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index e5bbe9a..405f8da 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -59,7 +59,6 @@ struct iommu_cmd {
 
 static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
 			     struct unity_map_entry *e);
-static struct dma_ops_domain *find_protection_domain(u16 devid);
 static u64 *alloc_pte(struct protection_domain *domain,
 		      unsigned long address, int end_lvl,
 		      u64 **pte_page, gfp_t gfp);
@@ -84,6 +83,34 @@ static inline u16 get_device_id(struct device *dev)
 	return calc_devid(pdev->bus->number, pdev->devfn);
 }
 
+/*
+ * In this function the list of preallocated protection domains is traversed to
+ * find the domain for a specific device
+ */
+static struct dma_ops_domain *find_protection_domain(u16 devid)
+{
+	struct dma_ops_domain *entry, *ret = NULL;
+	unsigned long flags;
+	u16 alias = amd_iommu_alias_table[devid];
+
+	if (list_empty(&iommu_pd_list))
+		return NULL;
+
+	spin_lock_irqsave(&iommu_pd_list_lock, flags);
+
+	list_for_each_entry(entry, &iommu_pd_list, list) {
+		if (entry->target_dev == devid ||
+		    entry->target_dev == alias) {
+			ret = entry;
+			break;
+		}
+	}
+
+	spin_unlock_irqrestore(&iommu_pd_list_lock, flags);
+
+	return ret;
+}
+
 #ifdef CONFIG_AMD_IOMMU_STATS
 
 /*
@@ -1456,34 +1483,6 @@ static bool check_device(struct device *dev)
 }
 
 /*
- * In this function the list of preallocated protection domains is traversed to
- * find the domain for a specific device
- */
-static struct dma_ops_domain *find_protection_domain(u16 devid)
-{
-	struct dma_ops_domain *entry, *ret = NULL;
-	unsigned long flags;
-	u16 alias = amd_iommu_alias_table[devid];
-
-	if (list_empty(&iommu_pd_list))
-		return NULL;
-
-	spin_lock_irqsave(&iommu_pd_list_lock, flags);
-
-	list_for_each_entry(entry, &iommu_pd_list, list) {
-		if (entry->target_dev == devid ||
-		    entry->target_dev == alias) {
-			ret = entry;
-			break;
-		}
-	}
-
-	spin_unlock_irqrestore(&iommu_pd_list_lock, flags);
-
-	return ret;
-}
-
-/*
  * In the dma_ops path we only have the struct device. This function
  * finds the corresponding IOMMU, the protection domain and the
  * requestor id for a given device.
-- 
1.6.5.3

[PATCH 22/34] x86/amd-iommu: Use get_device_id and check_device where appropriate

[Joerg Roedel]

The logic of these two functions is reimplemented (at least
in parts) in places in the code. This patch removes these
code duplications and uses the functions instead. As a side
effect it moves check_device() to the helper function code
section.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |  110 +++++++++++++++++++------------------------
 1 files changed, 49 insertions(+), 61 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 405f8da..d10195b 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -111,6 +111,33 @@ static struct dma_ops_domain *find_protection_domain(u16 devid)
 	return ret;
 }
 
+/*
+ * This function checks if the driver got a valid device from the caller to
+ * avoid dereferencing invalid pointers.
+ */
+static bool check_device(struct device *dev)
+{
+	u16 devid;
+
+	if (!dev || !dev->dma_mask)
+		return false;
+
+	/* No device or no PCI device */
+	if (!dev || dev->bus != &pci_bus_type)
+		return false;
+
+	devid = get_device_id(dev);
+
+	/* Out of our scope? */
+	if (devid > amd_iommu_last_bdf)
+		return false;
+
+	if (amd_iommu_rlookup_table[devid] == NULL)
+		return false;
+
+	return true;
+}
+
 #ifdef CONFIG_AMD_IOMMU_STATS
 
 /*
@@ -1386,22 +1413,17 @@ static int device_change_notifier(struct notifier_block *nb,
 				  unsigned long action, void *data)
 {
 	struct device *dev = data;
-	struct pci_dev *pdev = to_pci_dev(dev);
-	u16 devid = calc_devid(pdev->bus->number, pdev->devfn);
+	u16 devid;
 	struct protection_domain *domain;
 	struct dma_ops_domain *dma_domain;
 	struct amd_iommu *iommu;
 	unsigned long flags;
 
-	if (devid > amd_iommu_last_bdf)
-		goto out;
-
-	devid = amd_iommu_alias_table[devid];
-
-	iommu = amd_iommu_rlookup_table[devid];
-	if (iommu == NULL)
-		goto out;
+	if (!check_device(dev))
+		return 0;
 
+	devid  = get_device_id(dev);
+	iommu  = amd_iommu_rlookup_table[devid];
 	domain = domain_for_device(dev);
 
 	if (domain && !dma_ops_domain(domain))
@@ -1453,36 +1475,6 @@ static struct notifier_block device_nb = {
  *****************************************************************************/
 
 /*
- * This function checks if the driver got a valid device from the caller to
- * avoid dereferencing invalid pointers.
- */
-static bool check_device(struct device *dev)
-{
-	u16 bdf;
-	struct pci_dev *pcidev;
-
-	if (!dev || !dev->dma_mask)
-		return false;
-
-	/* No device or no PCI device */
-	if (!dev || dev->bus != &pci_bus_type)
-		return false;
-
-	pcidev = to_pci_dev(dev);
-
-	bdf = calc_devid(pcidev->bus->number, pcidev->devfn);
-
-	/* Out of our scope? */
-	if (bdf > amd_iommu_last_bdf)
-		return false;
-
-	if (amd_iommu_rlookup_table[bdf] == NULL)
-		return false;
-
-	return true;
-}
-
-/*
  * In the dma_ops path we only have the struct device. This function
  * finds the corresponding IOMMU, the protection domain and the
  * requestor id for a given device.
@@ -2094,15 +2086,20 @@ static void prealloc_protection_domains(void)
 	struct pci_dev *dev = NULL;
 	struct dma_ops_domain *dma_dom;
 	struct amd_iommu *iommu;
-	u16 devid, __devid;
+	u16 devid;
 
 	while ((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) {
-		__devid = devid = calc_devid(dev->bus->number, dev->devfn);
-		if (devid > amd_iommu_last_bdf)
+
+		/* Do we handle this device? */
+		if (!check_device(&dev->dev))
 			continue;
-		devid = amd_iommu_alias_table[devid];
+
+		/* Is there already any domain for it? */
 		if (domain_for_device(&dev->dev))
 			continue;
+
+		devid = get_device_id(&dev->dev);
+
 		iommu = amd_iommu_rlookup_table[devid];
 		if (!iommu)
 			continue;
@@ -2294,17 +2291,14 @@ static void amd_iommu_detach_device(struct iommu_domain *dom,
 				    struct device *dev)
 {
 	struct amd_iommu *iommu;
-	struct pci_dev *pdev;
 	u16 devid;
 
-	if (dev->bus != &pci_bus_type)
+	if (!check_device(dev))
 		return;
 
-	pdev = to_pci_dev(dev);
-
-	devid = calc_devid(pdev->bus->number, pdev->devfn);
+	devid = get_device_id(dev);
 
-	if (devid > 0)
+	if (amd_iommu_pd_table[devid] != NULL)
 		detach_device(dev);
 
 	iommu = amd_iommu_rlookup_table[devid];
@@ -2321,20 +2315,13 @@ static int amd_iommu_attach_device(struct iommu_domain *dom,
 	struct protection_domain *domain = dom->priv;
 	struct protection_domain *old_domain;
 	struct amd_iommu *iommu;
-	struct pci_dev *pdev;
 	int ret;
 	u16 devid;
 
-	if (dev->bus != &pci_bus_type)
+	if (!check_device(dev))
 		return -EINVAL;
 
-	pdev = to_pci_dev(dev);
-
-	devid = calc_devid(pdev->bus->number, pdev->devfn);
-
-	if (devid >= amd_iommu_last_bdf ||
-			devid != amd_iommu_alias_table[devid])
-		return -EINVAL;
+	devid = get_device_id(dev);
 
 	iommu = amd_iommu_rlookup_table[devid];
 	if (!iommu)
@@ -2458,10 +2445,11 @@ int __init amd_iommu_init_passthrough(void)
 
 	while ((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) {
 
-		devid = calc_devid(dev->bus->number, dev->devfn);
-		if (devid > amd_iommu_last_bdf)
+		if (!check_device(&dev->dev))
 			continue;
 
+		devid = get_device_id(&dev->dev);
+
 		iommu = amd_iommu_rlookup_table[devid];
 		if (!iommu)
 			continue;
-- 
1.6.5.3

[PATCH 23/34] x86/amd-iommu: Remove iommu parameter from dma_ops_domain_alloc

[Joerg Roedel]

This function doesn't use the parameter anymore so it can be
removed.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   12 ++++--------
 1 files changed, 4 insertions(+), 8 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index d10195b..17e83ec 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1188,7 +1188,7 @@ static void dma_ops_domain_free(struct dma_ops_domain *dom)
  * It also intializes the page table and the address allocator data
  * structures required for the dma_ops interface
  */
-static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu)
+static struct dma_ops_domain *dma_ops_domain_alloc(void)
 {
 	struct dma_ops_domain *dma_dom;
 
@@ -1443,7 +1443,7 @@ static int device_change_notifier(struct notifier_block *nb,
 		dma_domain = find_protection_domain(devid);
 		if (dma_domain)
 			goto out;
-		dma_domain = dma_ops_domain_alloc(iommu);
+		dma_domain = dma_ops_domain_alloc();
 		if (!dma_domain)
 			goto out;
 		dma_domain->target_dev = devid;
@@ -2085,7 +2085,6 @@ static void prealloc_protection_domains(void)
 {
 	struct pci_dev *dev = NULL;
 	struct dma_ops_domain *dma_dom;
-	struct amd_iommu *iommu;
 	u16 devid;
 
 	while ((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) {
@@ -2100,10 +2099,7 @@ static void prealloc_protection_domains(void)
 
 		devid = get_device_id(&dev->dev);
 
-		iommu = amd_iommu_rlookup_table[devid];
-		if (!iommu)
-			continue;
-		dma_dom = dma_ops_domain_alloc(iommu);
+		dma_dom = dma_ops_domain_alloc();
 		if (!dma_dom)
 			continue;
 		init_unity_mappings_for_device(dma_dom, devid);
@@ -2139,7 +2135,7 @@ int __init amd_iommu_init_dma_ops(void)
 	 * protection domain will be assigned to the default one.
 	 */
 	for_each_iommu(iommu) {
-		iommu->default_dom = dma_ops_domain_alloc(iommu);
+		iommu->default_dom = dma_ops_domain_alloc();
 		if (iommu->default_dom == NULL)
 			return -ENOMEM;
 		iommu->default_dom->domain.flags |= PD_DEFAULT_MASK;
-- 
1.6.5.3

[PATCH 24/34] x86/amd-iommu: Move some pte allocation functions in the right section

[Joerg Roedel]

This patch moves alloc_pte() and fetch_pte() into the page
table handling code section so that the forward declarations
for them could be removed.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |  193 +++++++++++++++++++++----------------------
 1 files changed, 94 insertions(+), 99 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 17e83ec..90b3650 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -59,15 +59,10 @@ struct iommu_cmd {
 
 static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
 			     struct unity_map_entry *e);
-static u64 *alloc_pte(struct protection_domain *domain,
-		      unsigned long address, int end_lvl,
-		      u64 **pte_page, gfp_t gfp);
 static void dma_ops_reserve_addresses(struct dma_ops_domain *dom,
 				      unsigned long start_page,
 				      unsigned int pages);
 static void reset_iommu_command_buffer(struct amd_iommu *iommu);
-static u64 *fetch_pte(struct protection_domain *domain,
-		      unsigned long address, int map_size);
 static void update_domain(struct protection_domain *domain);
 
 /****************************************************************************
@@ -665,6 +660,100 @@ void amd_iommu_flush_all_devices(void)
  ****************************************************************************/
 
 /*
+ * This function is used to add another level to an IO page table. Adding
+ * another level increases the size of the address space by 9 bits to a size up
+ * to 64 bits.
+ */
+static bool increase_address_space(struct protection_domain *domain,
+				   gfp_t gfp)
+{
+	u64 *pte;
+
+	if (domain->mode == PAGE_MODE_6_LEVEL)
+		/* address space already 64 bit large */
+		return false;
+
+	pte = (void *)get_zeroed_page(gfp);
+	if (!pte)
+		return false;
+
+	*pte             = PM_LEVEL_PDE(domain->mode,
+					virt_to_phys(domain->pt_root));
+	domain->pt_root  = pte;
+	domain->mode    += 1;
+	domain->updated  = true;
+
+	return true;
+}
+
+static u64 *alloc_pte(struct protection_domain *domain,
+		      unsigned long address,
+		      int end_lvl,
+		      u64 **pte_page,
+		      gfp_t gfp)
+{
+	u64 *pte, *page;
+	int level;
+
+	while (address > PM_LEVEL_SIZE(domain->mode))
+		increase_address_space(domain, gfp);
+
+	level =  domain->mode - 1;
+	pte   = &domain->pt_root[PM_LEVEL_INDEX(level, address)];
+
+	while (level > end_lvl) {
+		if (!IOMMU_PTE_PRESENT(*pte)) {
+			page = (u64 *)get_zeroed_page(gfp);
+			if (!page)
+				return NULL;
+			*pte = PM_LEVEL_PDE(level, virt_to_phys(page));
+		}
+
+		level -= 1;
+
+		pte = IOMMU_PTE_PAGE(*pte);
+
+		if (pte_page && level == end_lvl)
+			*pte_page = pte;
+
+		pte = &pte[PM_LEVEL_INDEX(level, address)];
+	}
+
+	return pte;
+}
+
+/*
+ * This function checks if there is a PTE for a given dma address. If
+ * there is one, it returns the pointer to it.
+ */
+static u64 *fetch_pte(struct protection_domain *domain,
+		      unsigned long address, int map_size)
+{
+	int level;
+	u64 *pte;
+
+	level =  domain->mode - 1;
+	pte   = &domain->pt_root[PM_LEVEL_INDEX(level, address)];
+
+	while (level > map_size) {
+		if (!IOMMU_PTE_PRESENT(*pte))
+			return NULL;
+
+		level -= 1;
+
+		pte = IOMMU_PTE_PAGE(*pte);
+		pte = &pte[PM_LEVEL_INDEX(level, address)];
+
+		if ((PM_PTE_LEVEL(*pte) == 0) && level != map_size) {
+			pte = NULL;
+			break;
+		}
+	}
+
+	return pte;
+}
+
+/*
  * Generic mapping functions. It maps a physical address into a DMA
  * address space. It allocates the page table pages if necessary.
  * In the future it can be extended to a generic mapping function
@@ -820,37 +909,6 @@ static int init_unity_mappings_for_device(struct dma_ops_domain *dma_dom,
  */
 
 /*
- * This function checks if there is a PTE for a given dma address. If
- * there is one, it returns the pointer to it.
- */
-static u64 *fetch_pte(struct protection_domain *domain,
-		      unsigned long address, int map_size)
-{
-	int level;
-	u64 *pte;
-
-	level =  domain->mode - 1;
-	pte   = &domain->pt_root[PM_LEVEL_INDEX(level, address)];
-
-	while (level > map_size) {
-		if (!IOMMU_PTE_PRESENT(*pte))
-			return NULL;
-
-		level -= 1;
-
-		pte = IOMMU_PTE_PAGE(*pte);
-		pte = &pte[PM_LEVEL_INDEX(level, address)];
-
-		if ((PM_PTE_LEVEL(*pte) == 0) && level != map_size) {
-			pte = NULL;
-			break;
-		}
-	}
-
-	return pte;
-}
-
-/*
  * This function is used to add a new aperture range to an existing
  * aperture in case of dma_ops domain allocation or address allocation
  * failure.
@@ -1535,69 +1593,6 @@ static void update_domain(struct protection_domain *domain)
 }
 
 /*
- * This function is used to add another level to an IO page table. Adding
- * another level increases the size of the address space by 9 bits to a size up
- * to 64 bits.
- */
-static bool increase_address_space(struct protection_domain *domain,
-				   gfp_t gfp)
-{
-	u64 *pte;
-
-	if (domain->mode == PAGE_MODE_6_LEVEL)
-		/* address space already 64 bit large */
-		return false;
-
-	pte = (void *)get_zeroed_page(gfp);
-	if (!pte)
-		return false;
-
-	*pte             = PM_LEVEL_PDE(domain->mode,
-					virt_to_phys(domain->pt_root));
-	domain->pt_root  = pte;
-	domain->mode    += 1;
-	domain->updated  = true;
-
-	return true;
-}
-
-static u64 *alloc_pte(struct protection_domain *domain,
-		      unsigned long address,
-		      int end_lvl,
-		      u64 **pte_page,
-		      gfp_t gfp)
-{
-	u64 *pte, *page;
-	int level;
-
-	while (address > PM_LEVEL_SIZE(domain->mode))
-		increase_address_space(domain, gfp);
-
-	level =  domain->mode - 1;
-	pte   = &domain->pt_root[PM_LEVEL_INDEX(level, address)];
-
-	while (level > end_lvl) {
-		if (!IOMMU_PTE_PRESENT(*pte)) {
-			page = (u64 *)get_zeroed_page(gfp);
-			if (!page)
-				return NULL;
-			*pte = PM_LEVEL_PDE(level, virt_to_phys(page));
-		}
-
-		level -= 1;
-
-		pte = IOMMU_PTE_PAGE(*pte);
-
-		if (pte_page && level == end_lvl)
-			*pte_page = pte;
-
-		pte = &pte[PM_LEVEL_INDEX(level, address)];
-	}
-
-	return pte;
-}
-
-/*
  * This function fetches the PTE for a given address in the aperture
  */
 static u64* dma_ops_get_pte(struct dma_ops_domain *dom,
-- 
1.6.5.3

[PATCH 25/34] x86/amd-iommu: Rearrange dma_ops related functions

[Joerg Roedel]

This patch rearranges two dma_ops related functions so that
their forward declarations are not longer necessary.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   89 ++++++++++++++++++++----------------------
 1 files changed, 42 insertions(+), 47 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 90b3650..14b60c0 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -57,11 +57,6 @@ struct iommu_cmd {
 	u32 data[4];
 };
 
-static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
-			     struct unity_map_entry *e);
-static void dma_ops_reserve_addresses(struct dma_ops_domain *dom,
-				      unsigned long start_page,
-				      unsigned int pages);
 static void reset_iommu_command_buffer(struct amd_iommu *iommu);
 static void update_domain(struct protection_domain *domain);
 
@@ -823,28 +818,6 @@ static int iommu_for_unity_map(struct amd_iommu *iommu,
 }
 
 /*
- * Init the unity mappings for a specific IOMMU in the system
- *
- * Basically iterates over all unity mapping entries and applies them to
- * the default domain DMA of that IOMMU if necessary.
- */
-static int iommu_init_unity_mappings(struct amd_iommu *iommu)
-{
-	struct unity_map_entry *entry;
-	int ret;
-
-	list_for_each_entry(entry, &amd_iommu_unity_map, list) {
-		if (!iommu_for_unity_map(iommu, entry))
-			continue;
-		ret = dma_ops_unity_map(iommu->default_dom, entry);
-		if (ret)
-			return ret;
-	}
-
-	return 0;
-}
-
-/*
  * This function actually applies the mapping to the page table of the
  * dma_ops domain.
  */
@@ -873,6 +846,28 @@ static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
 }
 
 /*
+ * Init the unity mappings for a specific IOMMU in the system
+ *
+ * Basically iterates over all unity mapping entries and applies them to
+ * the default domain DMA of that IOMMU if necessary.
+ */
+static int iommu_init_unity_mappings(struct amd_iommu *iommu)
+{
+	struct unity_map_entry *entry;
+	int ret;
+
+	list_for_each_entry(entry, &amd_iommu_unity_map, list) {
+		if (!iommu_for_unity_map(iommu, entry))
+			continue;
+		ret = dma_ops_unity_map(iommu->default_dom, entry);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+/*
  * Inits the unity mappings required for a specific device
  */
 static int init_unity_mappings_for_device(struct dma_ops_domain *dma_dom,
@@ -909,6 +904,26 @@ static int init_unity_mappings_for_device(struct dma_ops_domain *dma_dom,
  */
 
 /*
+ * Used to reserve address ranges in the aperture (e.g. for exclusion
+ * ranges.
+ */
+static void dma_ops_reserve_addresses(struct dma_ops_domain *dom,
+				      unsigned long start_page,
+				      unsigned int pages)
+{
+	unsigned int i, last_page = dom->aperture_size >> PAGE_SHIFT;
+
+	if (start_page + pages > last_page)
+		pages = last_page - start_page;
+
+	for (i = start_page; i < start_page + pages; ++i) {
+		int index = i / APERTURE_RANGE_PAGES;
+		int page  = i % APERTURE_RANGE_PAGES;
+		__set_bit(page, dom->aperture[index]->bitmap);
+	}
+}
+
+/*
  * This function is used to add a new aperture range to an existing
  * aperture in case of dma_ops domain allocation or address allocation
  * failure.
@@ -1166,26 +1181,6 @@ static void domain_id_free(int id)
 	write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
 }
 
-/*
- * Used to reserve address ranges in the aperture (e.g. for exclusion
- * ranges.
- */
-static void dma_ops_reserve_addresses(struct dma_ops_domain *dom,
-				      unsigned long start_page,
-				      unsigned int pages)
-{
-	unsigned int i, last_page = dom->aperture_size >> PAGE_SHIFT;
-
-	if (start_page + pages > last_page)
-		pages = last_page - start_page;
-
-	for (i = start_page; i < start_page + pages; ++i) {
-		int index = i / APERTURE_RANGE_PAGES;
-		int page  = i % APERTURE_RANGE_PAGES;
-		__set_bit(page, dom->aperture[index]->bitmap);
-	}
-}
-
 static void free_pagetable(struct protection_domain *domain)
 {
 	int i, j;
-- 
1.6.5.3

[PATCH 26/34] x86/amd-iommu: Remove support for domain sharing

[Joerg Roedel]

This patch makes device isolation mandatory and removes
support for the amd_iommu=share option. This simplifies the
code in several places.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu_types.h |    3 ---
 arch/x86/kernel/amd_iommu.c            |   10 ++--------
 arch/x86/kernel/amd_iommu_init.c       |   17 -----------------
 3 files changed, 2 insertions(+), 28 deletions(-)

diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index 4899f78..02b6a0f 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -451,9 +451,6 @@ extern struct protection_domain **amd_iommu_pd_table;
 /* allocation bitmap for domain ids */
 extern unsigned long *amd_iommu_pd_alloc_bitmap;
 
-/* will be 1 if device isolation is enabled */
-extern bool amd_iommu_isolate;
-
 /*
  * If true, the addresses will be flushed on unmap time, not when
  * they are reused
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 14b60c0..ed58a16 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -148,7 +148,6 @@ DECLARE_STATS_COUNTER(alloced_io_mem);
 DECLARE_STATS_COUNTER(total_map_requests);
 
 static struct dentry *stats_dir;
-static struct dentry *de_isolate;
 static struct dentry *de_fflush;
 
 static void amd_iommu_stats_add(struct __iommu_counter *cnt)
@@ -166,9 +165,6 @@ static void amd_iommu_stats_init(void)
 	if (stats_dir == NULL)
 		return;
 
-	de_isolate = debugfs_create_bool("isolation", 0444, stats_dir,
-					 (u32 *)&amd_iommu_isolate);
-
 	de_fflush  = debugfs_create_bool("fullflush", 0444, stats_dir,
 					 (u32 *)&amd_iommu_unmap_flush);
 
@@ -2135,11 +2131,9 @@ int __init amd_iommu_init_dma_ops(void)
 	}
 
 	/*
-	 * If device isolation is enabled, pre-allocate the protection
-	 * domains for each device.
+	 * Pre-allocate the protection domains for each device.
 	 */
-	if (amd_iommu_isolate)
-		prealloc_protection_domains();
+	prealloc_protection_domains();
 
 	iommu_detected = 1;
 	swiotlb = 0;
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index fbe4c3c..fe1686f 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -125,13 +125,6 @@ u16 amd_iommu_last_bdf;			/* largest PCI device id we have
 					   to handle */
 LIST_HEAD(amd_iommu_unity_map);		/* a list of required unity mappings
 					   we find in ACPI */
-#ifdef CONFIG_IOMMU_STRESS
-bool amd_iommu_isolate = false;
-#else
-bool amd_iommu_isolate = true;		/* if true, device isolation is
-					   enabled */
-#endif
-
 bool amd_iommu_unmap_flush;		/* if true, flush on every unmap */
 
 LIST_HEAD(amd_iommu_list);		/* list of all AMD IOMMUs in the
@@ -1308,12 +1301,6 @@ static int __init amd_iommu_init(void)
 	if (iommu_pass_through)
 		goto out;
 
-	printk(KERN_INFO "AMD-Vi: device isolation ");
-	if (amd_iommu_isolate)
-		printk("enabled\n");
-	else
-		printk("disabled\n");
-
 	if (amd_iommu_unmap_flush)
 		printk(KERN_INFO "AMD-Vi: IO/TLB flush on unmap enabled\n");
 	else
@@ -1387,10 +1374,6 @@ static int __init parse_amd_iommu_dump(char *str)
 static int __init parse_amd_iommu_options(char *str)
 {
 	for (; *str; ++str) {
-		if (strncmp(str, "isolate", 7) == 0)
-			amd_iommu_isolate = true;
-		if (strncmp(str, "share", 5) == 0)
-			amd_iommu_isolate = false;
 		if (strncmp(str, "fullflush", 9) == 0)
 			amd_iommu_unmap_flush = true;
 	}
-- 
1.6.5.3

[PATCH 27/34] x86/amd-iommu: Use dev->arch->iommu to store iommu related information

[Joerg Roedel]

This patch changes IOMMU code to use dev->archdata->iommu to
store information about the alias device and the domain the
device is attached to.
This allows the driver to get rid of the amd_iommu_pd_table
in the future.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu_types.h |    8 +++
 arch/x86/include/asm/device.h          |    2 +-
 arch/x86/kernel/amd_iommu.c            |  109 +++++++++++++++++++++++++-------
 3 files changed, 95 insertions(+), 24 deletions(-)

diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index 02b6a0f..9eaa27b 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -248,6 +248,14 @@ struct protection_domain {
 };
 
 /*
+ * This struct contains device specific data for the IOMMU
+ */
+struct iommu_dev_data {
+	struct device *alias;		  /* The Alias Device */
+	struct protection_domain *domain; /* Domain the device is bound to */
+};
+
+/*
  * For dynamic growth the aperture size is split into ranges of 128MB of
  * DMA address space each. This struct represents one such range.
  */
diff --git a/arch/x86/include/asm/device.h b/arch/x86/include/asm/device.h
index cee34e9..029f230 100644
--- a/arch/x86/include/asm/device.h
+++ b/arch/x86/include/asm/device.h
@@ -8,7 +8,7 @@ struct dev_archdata {
 #ifdef CONFIG_X86_64
 struct dma_map_ops *dma_ops;
 #endif
-#ifdef CONFIG_DMAR
+#if defined(CONFIG_DMAR) || defined(CONFIG_AMD_IOMMU)
 	void *iommu; /* hook for IOMMU specific extension */
 #endif
 };
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index ed58a16..3214e88 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -73,6 +73,11 @@ static inline u16 get_device_id(struct device *dev)
 	return calc_devid(pdev->bus->number, pdev->devfn);
 }
 
+static struct iommu_dev_data *get_dev_data(struct device *dev)
+{
+	return dev->archdata.iommu;
+}
+
 /*
  * In this function the list of preallocated protection domains is traversed to
  * find the domain for a specific device
@@ -128,6 +133,35 @@ static bool check_device(struct device *dev)
 	return true;
 }
 
+static int iommu_init_device(struct device *dev)
+{
+	struct iommu_dev_data *dev_data;
+	struct pci_dev *pdev;
+	u16 devid, alias;
+
+	if (dev->archdata.iommu)
+		return 0;
+
+	dev_data = kzalloc(sizeof(*dev_data), GFP_KERNEL);
+	if (!dev_data)
+		return -ENOMEM;
+
+	devid = get_device_id(dev);
+	alias = amd_iommu_alias_table[devid];
+	pdev = pci_get_bus_and_slot(PCI_BUS(alias), alias & 0xff);
+	if (pdev)
+		dev_data->alias = &pdev->dev;
+
+	dev->archdata.iommu = dev_data;
+
+
+	return 0;
+}
+
+static void iommu_uninit_device(struct device *dev)
+{
+	kfree(dev->archdata.iommu);
+}
 #ifdef CONFIG_AMD_IOMMU_STATS
 
 /*
@@ -1346,28 +1380,39 @@ static void clear_dte_entry(u16 devid)
 static int __attach_device(struct device *dev,
 			   struct protection_domain *domain)
 {
-	u16 devid = get_device_id(dev);
-	u16 alias = amd_iommu_alias_table[devid];
+	struct iommu_dev_data *dev_data, *alias_data;
+	u16 devid, alias;
+
+	devid      = get_device_id(dev);
+	alias      = amd_iommu_alias_table[devid];
+	dev_data   = get_dev_data(dev);
+	alias_data = get_dev_data(dev_data->alias);
+	if (!alias_data)
+		return -EINVAL;
 
 	/* lock domain */
 	spin_lock(&domain->lock);
 
 	/* Some sanity checks */
-	if (amd_iommu_pd_table[alias] != NULL &&
-	    amd_iommu_pd_table[alias] != domain)
+	if (alias_data->domain != NULL &&
+	    alias_data->domain != domain)
 		return -EBUSY;
 
-	if (amd_iommu_pd_table[devid] != NULL &&
-	    amd_iommu_pd_table[devid] != domain)
+	if (dev_data->domain != NULL &&
+	    dev_data->domain != domain)
 		return -EBUSY;
 
 	/* Do real assignment */
 	if (alias != devid &&
-	    amd_iommu_pd_table[alias] == NULL)
+	    alias_data->domain == NULL) {
+		alias_data->domain = domain;
 		set_dte_entry(alias, domain);
+	}
 
-	if (amd_iommu_pd_table[devid] == NULL)
+	if (dev_data->domain == NULL) {
+		dev_data->domain = domain;
 		set_dte_entry(devid, domain);
+	}
 
 	/* ready */
 	spin_unlock(&domain->lock);
@@ -1406,10 +1451,12 @@ static void __detach_device(struct device *dev)
 {
 	u16 devid = get_device_id(dev);
 	struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
+	struct iommu_dev_data *dev_data = get_dev_data(dev);
 
 	BUG_ON(!iommu);
 
 	clear_dte_entry(devid);
+	dev_data->domain = NULL;
 
 	/*
 	 * If we run in passthrough mode the device must be assigned to the
@@ -1439,18 +1486,23 @@ static void detach_device(struct device *dev)
 static struct protection_domain *domain_for_device(struct device *dev)
 {
 	struct protection_domain *dom;
+	struct iommu_dev_data *dev_data, *alias_data;
 	unsigned long flags;
 	u16 devid, alias;
 
-	devid = get_device_id(dev);
-	alias = amd_iommu_alias_table[devid];
+	devid      = get_device_id(dev);
+	alias      = amd_iommu_alias_table[devid];
+	dev_data   = get_dev_data(dev);
+	alias_data = get_dev_data(dev_data->alias);
+	if (!alias_data)
+		return NULL;
 
 	read_lock_irqsave(&amd_iommu_devtable_lock, flags);
-	dom = amd_iommu_pd_table[devid];
+	dom = dev_data->domain;
 	if (dom == NULL &&
-	    amd_iommu_pd_table[alias] != NULL) {
-		__attach_device(dev, amd_iommu_pd_table[alias]);
-		dom = amd_iommu_pd_table[devid];
+	    alias_data->domain != NULL) {
+		__attach_device(dev, alias_data->domain);
+		dom = alias_data->domain;
 	}
 
 	read_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
@@ -1473,14 +1525,12 @@ static int device_change_notifier(struct notifier_block *nb,
 
 	devid  = get_device_id(dev);
 	iommu  = amd_iommu_rlookup_table[devid];
-	domain = domain_for_device(dev);
-
-	if (domain && !dma_ops_domain(domain))
-		WARN_ONCE(1, "AMD IOMMU WARNING: device %s already bound "
-			  "to a non-dma-ops domain\n", dev_name(dev));
 
 	switch (action) {
 	case BUS_NOTIFY_UNBOUND_DRIVER:
+
+		domain = domain_for_device(dev);
+
 		if (!domain)
 			goto out;
 		if (iommu_pass_through)
@@ -1488,6 +1538,11 @@ static int device_change_notifier(struct notifier_block *nb,
 		detach_device(dev);
 		break;
 	case BUS_NOTIFY_ADD_DEVICE:
+
+		iommu_init_device(dev);
+
+		domain = domain_for_device(dev);
+
 		/* allocate a protection domain if a device is added */
 		dma_domain = find_protection_domain(devid);
 		if (dma_domain)
@@ -1502,6 +1557,10 @@ static int device_change_notifier(struct notifier_block *nb,
 		spin_unlock_irqrestore(&iommu_pd_list_lock, flags);
 
 		break;
+	case BUS_NOTIFY_DEL_DEVICE:
+
+		iommu_uninit_device(dev);
+
 	default:
 		goto out;
 	}
@@ -2079,6 +2138,8 @@ static void prealloc_protection_domains(void)
 		if (!check_device(&dev->dev))
 			continue;
 
+		iommu_init_device(&dev->dev);
+
 		/* Is there already any domain for it? */
 		if (domain_for_device(&dev->dev))
 			continue;
@@ -2270,6 +2331,7 @@ static void amd_iommu_domain_destroy(struct iommu_domain *dom)
 static void amd_iommu_detach_device(struct iommu_domain *dom,
 				    struct device *dev)
 {
+	struct iommu_dev_data *dev_data = dev->archdata.iommu;
 	struct amd_iommu *iommu;
 	u16 devid;
 
@@ -2278,7 +2340,7 @@ static void amd_iommu_detach_device(struct iommu_domain *dom,
 
 	devid = get_device_id(dev);
 
-	if (amd_iommu_pd_table[devid] != NULL)
+	if (dev_data->domain != NULL)
 		detach_device(dev);
 
 	iommu = amd_iommu_rlookup_table[devid];
@@ -2293,7 +2355,7 @@ static int amd_iommu_attach_device(struct iommu_domain *dom,
 				   struct device *dev)
 {
 	struct protection_domain *domain = dom->priv;
-	struct protection_domain *old_domain;
+	struct iommu_dev_data *dev_data;
 	struct amd_iommu *iommu;
 	int ret;
 	u16 devid;
@@ -2301,14 +2363,15 @@ static int amd_iommu_attach_device(struct iommu_domain *dom,
 	if (!check_device(dev))
 		return -EINVAL;
 
+	dev_data = dev->archdata.iommu;
+
 	devid = get_device_id(dev);
 
 	iommu = amd_iommu_rlookup_table[devid];
 	if (!iommu)
 		return -EINVAL;
 
-	old_domain = amd_iommu_pd_table[devid];
-	if (old_domain)
+	if (dev_data->domain)
 		detach_device(dev);
 
 	ret = attach_device(dev, domain);
-- 
1.6.5.3

[PATCH 28/34] x86/amd-iommu: Add device bind reference counting

[Joerg Roedel]

This patch adds a reference count to each device to count
how often the device was bound to that domain. This is
important for single devices that act as an alias for a
number of others. These devices must stay bound to their
domains until all devices that alias to it are unbound from
the same domain.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu_types.h |    1 +
 arch/x86/kernel/amd_iommu.c            |   37 +++++++++++++++++++++++++-------
 2 files changed, 30 insertions(+), 8 deletions(-)

diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index 9eaa27b..434e90e 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -253,6 +253,7 @@ struct protection_domain {
 struct iommu_dev_data {
 	struct device *alias;		  /* The Alias Device */
 	struct protection_domain *domain; /* Domain the device is bound to */
+	atomic_t bind;			  /* Domain attach reverent count */
 };
 
 /*
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 3214e88..f5db7d5 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -152,6 +152,8 @@ static int iommu_init_device(struct device *dev)
 	if (pdev)
 		dev_data->alias = &pdev->dev;
 
+	atomic_set(&dev_data->bind, 0);
+
 	dev->archdata.iommu = dev_data;
 
 
@@ -1403,10 +1405,13 @@ static int __attach_device(struct device *dev,
 		return -EBUSY;
 
 	/* Do real assignment */
-	if (alias != devid &&
-	    alias_data->domain == NULL) {
-		alias_data->domain = domain;
-		set_dte_entry(alias, domain);
+	if (alias != devid) {
+		if (alias_data->domain == NULL) {
+			alias_data->domain = domain;
+			set_dte_entry(alias, domain);
+		}
+
+		atomic_inc(&alias_data->bind);
 	}
 
 	if (dev_data->domain == NULL) {
@@ -1414,6 +1419,8 @@ static int __attach_device(struct device *dev,
 		set_dte_entry(devid, domain);
 	}
 
+	atomic_inc(&dev_data->bind);
+
 	/* ready */
 	spin_unlock(&domain->lock);
 
@@ -1449,20 +1456,34 @@ static int attach_device(struct device *dev,
  */
 static void __detach_device(struct device *dev)
 {
-	u16 devid = get_device_id(dev);
+	u16 devid = get_device_id(dev), alias;
 	struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
 	struct iommu_dev_data *dev_data = get_dev_data(dev);
+	struct iommu_dev_data *alias_data;
 
 	BUG_ON(!iommu);
 
-	clear_dte_entry(devid);
-	dev_data->domain = NULL;
+	devid = get_device_id(dev);
+	alias = get_device_id(dev_data->alias);
+
+	if (devid != alias) {
+		alias_data = get_dev_data(dev_data->alias);
+		if (atomic_dec_and_test(&alias_data->bind)) {
+			clear_dte_entry(alias);
+			alias_data->domain = NULL;
+		}
+	}
+
+	if (atomic_dec_and_test(&dev_data->bind)) {
+		clear_dte_entry(devid);
+		dev_data->domain = NULL;
+	}
 
 	/*
 	 * If we run in passthrough mode the device must be assigned to the
 	 * passthrough domain if it is detached from any other domain
 	 */
-	if (iommu_pass_through)
+	if (iommu_pass_through && dev_data->domain == NULL)
 		__attach_device(dev, pt_domain);
 }
 
-- 
1.6.5.3

[PATCH 29/34] x86/amd-iommu: Keep devices per domain in a list

[Joerg Roedel]

This patch introduces a list to each protection domain which
keeps all devices associated with the domain. This can be
used later to optimize certain functions and to completly
remove the amd_iommu_pd_table.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu_types.h |    2 ++
 arch/x86/kernel/amd_iommu.c            |   11 +++++++++++
 2 files changed, 13 insertions(+), 0 deletions(-)

diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index 434e90e..93953d1 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -235,6 +235,7 @@ extern bool amd_iommu_np_cache;
  */
 struct protection_domain {
 	struct list_head list;  /* for list of all protection domains */
+	struct list_head dev_list; /* List of all devices in this domain */
 	spinlock_t lock;	/* mostly used to lock the page table*/
 	u16 id;			/* the domain id written to the device table */
 	int mode;		/* paging mode (0-6 levels) */
@@ -251,6 +252,7 @@ struct protection_domain {
  * This struct contains device specific data for the IOMMU
  */
 struct iommu_dev_data {
+	struct list_head list;		  /* For domain->dev_list */
 	struct device *alias;		  /* The Alias Device */
 	struct protection_domain *domain; /* Domain the device is bound to */
 	atomic_t bind;			  /* Domain attach reverent count */
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index f5db7d5..530d608 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1286,6 +1286,7 @@ static struct dma_ops_domain *dma_ops_domain_alloc(void)
 	dma_dom->domain.id = domain_id_alloc();
 	if (dma_dom->domain.id == 0)
 		goto free_dma_dom;
+	INIT_LIST_HEAD(&dma_dom->domain.dev_list);
 	dma_dom->domain.mode = PAGE_MODE_2_LEVEL;
 	dma_dom->domain.pt_root = (void *)get_zeroed_page(GFP_KERNEL);
 	dma_dom->domain.flags = PD_DMA_OPS_MASK;
@@ -1408,6 +1409,7 @@ static int __attach_device(struct device *dev,
 	if (alias != devid) {
 		if (alias_data->domain == NULL) {
 			alias_data->domain = domain;
+			list_add(&alias_data->list, &domain->dev_list);
 			set_dte_entry(alias, domain);
 		}
 
@@ -1416,6 +1418,7 @@ static int __attach_device(struct device *dev,
 
 	if (dev_data->domain == NULL) {
 		dev_data->domain = domain;
+		list_add(&dev_data->list, &domain->dev_list);
 		set_dte_entry(devid, domain);
 	}
 
@@ -1460,6 +1463,7 @@ static void __detach_device(struct device *dev)
 	struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
 	struct iommu_dev_data *dev_data = get_dev_data(dev);
 	struct iommu_dev_data *alias_data;
+	unsigned long flags;
 
 	BUG_ON(!iommu);
 
@@ -1469,13 +1473,19 @@ static void __detach_device(struct device *dev)
 	if (devid != alias) {
 		alias_data = get_dev_data(dev_data->alias);
 		if (atomic_dec_and_test(&alias_data->bind)) {
+			spin_lock_irqsave(&alias_data->domain->lock, flags);
 			clear_dte_entry(alias);
+			list_del(&alias_data->list);
+			spin_unlock_irqrestore(&alias_data->domain->lock, flags);
 			alias_data->domain = NULL;
 		}
 	}
 
 	if (atomic_dec_and_test(&dev_data->bind)) {
+		spin_lock_irqsave(&dev_data->domain->lock, flags);
 		clear_dte_entry(devid);
+		list_del(&dev_data->list);
+		spin_unlock_irqrestore(&dev_data->domain->lock, flags);
 		dev_data->domain = NULL;
 	}
 
@@ -2294,6 +2304,7 @@ static struct protection_domain *protection_domain_alloc(void)
 	domain->id = domain_id_alloc();
 	if (!domain->id)
 		goto out_err;
+	INIT_LIST_HEAD(&domain->dev_list);
 
 	add_domain_to_list(domain);
 
-- 
1.6.5.3

[PATCH 30/34] x86/amd-iommu: Cleanup attach/detach_device code

[Joerg Roedel]

This patch cleans up the attach_device and detach_device
paths and fixes reference counting while at it.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |  102 ++++++++++++++++++++++++------------------
 1 files changed, 58 insertions(+), 44 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 530d608..e3363fd 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1329,7 +1329,6 @@ static bool dma_ops_domain(struct protection_domain *domain)
 
 static void set_dte_entry(u16 devid, struct protection_domain *domain)
 {
-	struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
 	u64 pte_root = virt_to_phys(domain->pt_root);
 
 	BUG_ON(amd_iommu_pd_table[devid] != NULL);
@@ -1344,18 +1343,11 @@ static void set_dte_entry(u16 devid, struct protection_domain *domain)
 
 	amd_iommu_pd_table[devid] = domain;
 
-	/* Do reference counting */
-	domain->dev_iommu[iommu->index] += 1;
-	domain->dev_cnt                 += 1;
-
-	/* Flush the changes DTE entry */
-	iommu_queue_inv_dev_entry(iommu, devid);
 }
 
 static void clear_dte_entry(u16 devid)
 {
 	struct protection_domain *domain = amd_iommu_pd_table[devid];
-	struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
 
 	BUG_ON(domain == NULL);
 
@@ -1368,11 +1360,51 @@ static void clear_dte_entry(u16 devid)
 	amd_iommu_dev_table[devid].data[2] = 0;
 
 	amd_iommu_apply_erratum_63(devid);
+}
+
+static void do_attach(struct device *dev, struct protection_domain *domain)
+{
+	struct iommu_dev_data *dev_data;
+	struct amd_iommu *iommu;
+	u16 devid;
+
+	devid    = get_device_id(dev);
+	iommu    = amd_iommu_rlookup_table[devid];
+	dev_data = get_dev_data(dev);
+
+	/* Update data structures */
+	dev_data->domain = domain;
+	list_add(&dev_data->list, &domain->dev_list);
+	set_dte_entry(devid, domain);
+
+	/* Do reference counting */
+	domain->dev_iommu[iommu->index] += 1;
+	domain->dev_cnt                 += 1;
+
+	/* Flush the DTE entry */
+	iommu_queue_inv_dev_entry(iommu, devid);
+}
+
+static void do_detach(struct device *dev)
+{
+	struct iommu_dev_data *dev_data;
+	struct amd_iommu *iommu;
+	u16 devid;
+
+	devid    = get_device_id(dev);
+	iommu    = amd_iommu_rlookup_table[devid];
+	dev_data = get_dev_data(dev);
 
 	/* decrease reference counters */
-	domain->dev_iommu[iommu->index] -= 1;
-	domain->dev_cnt                 -= 1;
+	dev_data->domain->dev_iommu[iommu->index] -= 1;
+	dev_data->domain->dev_cnt                 -= 1;
+
+	/* Update data structures */
+	dev_data->domain = NULL;
+	list_del(&dev_data->list);
+	clear_dte_entry(devid);
 
+	/* Flush the DTE entry */
 	iommu_queue_inv_dev_entry(iommu, devid);
 }
 
@@ -1384,12 +1416,10 @@ static int __attach_device(struct device *dev,
 			   struct protection_domain *domain)
 {
 	struct iommu_dev_data *dev_data, *alias_data;
-	u16 devid, alias;
 
-	devid      = get_device_id(dev);
-	alias      = amd_iommu_alias_table[devid];
 	dev_data   = get_dev_data(dev);
 	alias_data = get_dev_data(dev_data->alias);
+
 	if (!alias_data)
 		return -EINVAL;
 
@@ -1406,21 +1436,16 @@ static int __attach_device(struct device *dev,
 		return -EBUSY;
 
 	/* Do real assignment */
-	if (alias != devid) {
-		if (alias_data->domain == NULL) {
-			alias_data->domain = domain;
-			list_add(&alias_data->list, &domain->dev_list);
-			set_dte_entry(alias, domain);
-		}
+	if (dev_data->alias != dev) {
+		alias_data = get_dev_data(dev_data->alias);
+		if (alias_data->domain == NULL)
+			do_attach(dev_data->alias, domain);
 
 		atomic_inc(&alias_data->bind);
 	}
 
-	if (dev_data->domain == NULL) {
-		dev_data->domain = domain;
-		list_add(&dev_data->list, &domain->dev_list);
-		set_dte_entry(devid, domain);
-	}
+	if (dev_data->domain == NULL)
+		do_attach(dev, domain);
 
 	atomic_inc(&dev_data->bind);
 
@@ -1459,35 +1484,24 @@ static int attach_device(struct device *dev,
  */
 static void __detach_device(struct device *dev)
 {
-	u16 devid = get_device_id(dev), alias;
-	struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
 	struct iommu_dev_data *dev_data = get_dev_data(dev);
 	struct iommu_dev_data *alias_data;
 	unsigned long flags;
 
-	BUG_ON(!iommu);
+	BUG_ON(!dev_data->domain);
 
-	devid = get_device_id(dev);
-	alias = get_device_id(dev_data->alias);
+	spin_lock_irqsave(&dev_data->domain->lock, flags);
 
-	if (devid != alias) {
+	if (dev_data->alias != dev) {
 		alias_data = get_dev_data(dev_data->alias);
-		if (atomic_dec_and_test(&alias_data->bind)) {
-			spin_lock_irqsave(&alias_data->domain->lock, flags);
-			clear_dte_entry(alias);
-			list_del(&alias_data->list);
-			spin_unlock_irqrestore(&alias_data->domain->lock, flags);
-			alias_data->domain = NULL;
-		}
+		if (atomic_dec_and_test(&alias_data->bind))
+			do_detach(dev_data->alias);
 	}
 
-	if (atomic_dec_and_test(&dev_data->bind)) {
-		spin_lock_irqsave(&dev_data->domain->lock, flags);
-		clear_dte_entry(devid);
-		list_del(&dev_data->list);
-		spin_unlock_irqrestore(&dev_data->domain->lock, flags);
-		dev_data->domain = NULL;
-	}
+	if (atomic_dec_and_test(&dev_data->bind))
+		do_detach(dev);
+
+	spin_unlock_irqrestore(&dev_data->domain->lock, flags);
 
 	/*
 	 * If we run in passthrough mode the device must be assigned to the
-- 
1.6.5.3

[PATCH 31/34] x86/amd-iommu: Introduce iommu_flush_device() function

[Joerg Roedel]

This patch adds a function to flush a DTE entry for a given
struct device and replaces iommu_queue_inv_dev_entry calls
with this function where appropriate.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   19 +++++++++++++++----
 1 files changed, 15 insertions(+), 4 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index e3363fd..41c4ebe 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -494,6 +494,17 @@ static int iommu_queue_inv_dev_entry(struct amd_iommu *iommu, u16 devid)
 	return ret;
 }
 
+static int iommu_flush_device(struct device *dev)
+{
+	struct amd_iommu *iommu;
+	u16 devid;
+
+	devid = get_device_id(dev);
+	iommu = amd_iommu_rlookup_table[devid];
+
+	return iommu_queue_inv_dev_entry(iommu, devid);
+}
+
 static void __iommu_build_inv_iommu_pages(struct iommu_cmd *cmd, u64 address,
 					  u16 domid, int pde, int s)
 {
@@ -1382,7 +1393,7 @@ static void do_attach(struct device *dev, struct protection_domain *domain)
 	domain->dev_cnt                 += 1;
 
 	/* Flush the DTE entry */
-	iommu_queue_inv_dev_entry(iommu, devid);
+	iommu_flush_device(dev);
 }
 
 static void do_detach(struct device *dev)
@@ -1405,7 +1416,7 @@ static void do_detach(struct device *dev)
 	clear_dte_entry(devid);
 
 	/* Flush the DTE entry */
-	iommu_queue_inv_dev_entry(iommu, devid);
+	iommu_flush_device(dev);
 }
 
 /*
@@ -1610,7 +1621,7 @@ static int device_change_notifier(struct notifier_block *nb,
 		goto out;
 	}
 
-	iommu_queue_inv_dev_entry(iommu, devid);
+	iommu_flush_device(dev);
 	iommu_completion_wait(iommu);
 
 out:
@@ -2393,7 +2404,7 @@ static void amd_iommu_detach_device(struct iommu_domain *dom,
 	if (!iommu)
 		return;
 
-	iommu_queue_inv_dev_entry(iommu, devid);
+	iommu_flush_device(dev);
 	iommu_completion_wait(iommu);
 }
 
-- 
1.6.5.3

[PATCH 32/34] x86/amd-iommu: Cleanup DTE flushing code

[Joerg Roedel]

This patch cleans up the code to flush device table entries
in the IOMMU. With this chance the driver can get rid of the
iommu_queue_inv_dev_entry() function.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu_types.h |    1 +
 arch/x86/kernel/amd_iommu.c            |  100 +++++++++++---------------------
 2 files changed, 35 insertions(+), 66 deletions(-)

diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index 93953d1..f92d1b3 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -253,6 +253,7 @@ struct protection_domain {
  */
 struct iommu_dev_data {
 	struct list_head list;		  /* For domain->dev_list */
+	struct device *dev;		  /* Device this data belong to */
 	struct device *alias;		  /* The Alias Device */
 	struct protection_domain *domain; /* Domain the device is bound to */
 	atomic_t bind;			  /* Domain attach reverent count */
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 41c4ebe..0eafca5 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -146,6 +146,8 @@ static int iommu_init_device(struct device *dev)
 	if (!dev_data)
 		return -ENOMEM;
 
+	dev_data->dev = dev;
+
 	devid = get_device_id(dev);
 	alias = amd_iommu_alias_table[devid];
 	pdev = pci_get_bus_and_slot(PCI_BUS(alias), alias & 0xff);
@@ -478,31 +480,21 @@ static void iommu_flush_complete(struct protection_domain *domain)
 /*
  * Command send function for invalidating a device table entry
  */
-static int iommu_queue_inv_dev_entry(struct amd_iommu *iommu, u16 devid)
-{
-	struct iommu_cmd cmd;
-	int ret;
-
-	BUG_ON(iommu == NULL);
-
-	memset(&cmd, 0, sizeof(cmd));
-	CMD_SET_TYPE(&cmd, CMD_INV_DEV_ENTRY);
-	cmd.data[0] = devid;
-
-	ret = iommu_queue_command(iommu, &cmd);
-
-	return ret;
-}
-
 static int iommu_flush_device(struct device *dev)
 {
 	struct amd_iommu *iommu;
+	struct iommu_cmd cmd;
 	u16 devid;
 
 	devid = get_device_id(dev);
 	iommu = amd_iommu_rlookup_table[devid];
 
-	return iommu_queue_inv_dev_entry(iommu, devid);
+	/* Build command */
+	memset(&cmd, 0, sizeof(cmd));
+	CMD_SET_TYPE(&cmd, CMD_INV_DEV_ENTRY);
+	cmd.data[0] = devid;
+
+	return iommu_queue_command(iommu, &cmd);
 }
 
 static void __iommu_build_inv_iommu_pages(struct iommu_cmd *cmd, u64 address,
@@ -592,30 +584,43 @@ static void iommu_flush_tlb_pde(struct protection_domain *domain)
 	__iommu_flush_pages(domain, 0, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, 1);
 }
 
+
 /*
- * This function flushes all domains that have devices on the given IOMMU
+ * This function flushes the DTEs for all devices in domain
  */
-static void flush_all_domains_on_iommu(struct amd_iommu *iommu)
+static void iommu_flush_domain_devices(struct protection_domain *domain)
+{
+	struct iommu_dev_data *dev_data;
+	unsigned long flags;
+
+	spin_lock_irqsave(&domain->lock, flags);
+
+	list_for_each_entry(dev_data, &domain->dev_list, list)
+		iommu_flush_device(dev_data->dev);
+
+	spin_unlock_irqrestore(&domain->lock, flags);
+}
+
+static void iommu_flush_all_domain_devices(void)
 {
-	u64 address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS;
 	struct protection_domain *domain;
 	unsigned long flags;
 
 	spin_lock_irqsave(&amd_iommu_pd_lock, flags);
 
 	list_for_each_entry(domain, &amd_iommu_pd_list, list) {
-		if (domain->dev_iommu[iommu->index] == 0)
-			continue;
-
-		spin_lock(&domain->lock);
-		iommu_queue_inv_iommu_pages(iommu, address, domain->id, 1, 1);
+		iommu_flush_domain_devices(domain);
 		iommu_flush_complete(domain);
-		spin_unlock(&domain->lock);
 	}
 
 	spin_unlock_irqrestore(&amd_iommu_pd_lock, flags);
 }
 
+void amd_iommu_flush_all_devices(void)
+{
+	iommu_flush_all_domain_devices();
+}
+
 /*
  * This function uses heavy locking and may disable irqs for some time. But
  * this is no issue because it is only called during resume.
@@ -637,38 +642,6 @@ void amd_iommu_flush_all_domains(void)
 	spin_unlock_irqrestore(&amd_iommu_pd_lock, flags);
 }
 
-static void flush_all_devices_for_iommu(struct amd_iommu *iommu)
-{
-	int i;
-
-	for (i = 0; i <= amd_iommu_last_bdf; ++i) {
-		if (iommu != amd_iommu_rlookup_table[i])
-			continue;
-
-		iommu_queue_inv_dev_entry(iommu, i);
-		iommu_completion_wait(iommu);
-	}
-}
-
-static void flush_devices_by_domain(struct protection_domain *domain)
-{
-	struct amd_iommu *iommu;
-	int i;
-
-	for (i = 0; i <= amd_iommu_last_bdf; ++i) {
-		if ((domain == NULL && amd_iommu_pd_table[i] == NULL) ||
-		    (amd_iommu_pd_table[i] != domain))
-			continue;
-
-		iommu = amd_iommu_rlookup_table[i];
-		if (!iommu)
-			continue;
-
-		iommu_queue_inv_dev_entry(iommu, i);
-		iommu_completion_wait(iommu);
-	}
-}
-
 static void reset_iommu_command_buffer(struct amd_iommu *iommu)
 {
 	pr_err("AMD-Vi: Resetting IOMMU command buffer\n");
@@ -679,17 +652,12 @@ static void reset_iommu_command_buffer(struct amd_iommu *iommu)
 	iommu->reset_in_progress = true;
 
 	amd_iommu_reset_cmd_buffer(iommu);
-	flush_all_devices_for_iommu(iommu);
-	flush_all_domains_on_iommu(iommu);
+	amd_iommu_flush_all_devices();
+	amd_iommu_flush_all_domains();
 
 	iommu->reset_in_progress = false;
 }
 
-void amd_iommu_flush_all_devices(void)
-{
-	flush_devices_by_domain(NULL);
-}
-
 /****************************************************************************
  *
  * The functions below are used the create the page table mappings for
@@ -1692,7 +1660,7 @@ static void update_domain(struct protection_domain *domain)
 		return;
 
 	update_device_table(domain);
-	flush_devices_by_domain(domain);
+	iommu_flush_domain_devices(domain);
 	iommu_flush_tlb_pde(domain);
 
 	domain->updated = false;
-- 
1.6.5.3

[PATCH 33/34] x86/amd-iommu: Move reset_iommu_command_buffer out of locked code

[Joerg Roedel]

This patch removes the ugly contruct where the
iommu->lock must be released while before calling the
reset_iommu_command_buffer function.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/amd_iommu.c |   13 ++++++-------
 1 files changed, 6 insertions(+), 7 deletions(-)

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 0eafca5..b75fcd9 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -285,6 +285,7 @@ static void iommu_print_event(struct amd_iommu *iommu, void *__evt)
 		break;
 	case EVENT_TYPE_ILL_CMD:
 		printk("ILLEGAL_COMMAND_ERROR address=0x%016llx]\n", address);
+		iommu->reset_in_progress = true;
 		reset_iommu_command_buffer(iommu);
 		dump_command(address);
 		break;
@@ -407,11 +408,8 @@ static void __iommu_wait_for_completion(struct amd_iommu *iommu)
 	status &= ~MMIO_STATUS_COM_WAIT_INT_MASK;
 	writel(status, iommu->mmio_base + MMIO_STATUS_OFFSET);
 
-	if (unlikely(i == EXIT_LOOP_COUNT)) {
-		spin_unlock(&iommu->lock);
-		reset_iommu_command_buffer(iommu);
-		spin_lock(&iommu->lock);
-	}
+	if (unlikely(i == EXIT_LOOP_COUNT))
+		iommu->reset_in_progress = true;
 }
 
 /*
@@ -458,6 +456,9 @@ static int iommu_completion_wait(struct amd_iommu *iommu)
 out:
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
+	if (iommu->reset_in_progress)
+		reset_iommu_command_buffer(iommu);
+
 	return 0;
 }
 
@@ -649,8 +650,6 @@ static void reset_iommu_command_buffer(struct amd_iommu *iommu)
 	if (iommu->reset_in_progress)
 		panic("AMD-Vi: ILLEGAL_COMMAND_ERROR while resetting command buffer\n");
 
-	iommu->reset_in_progress = true;
-
 	amd_iommu_reset_cmd_buffer(iommu);
 	amd_iommu_flush_all_devices();
 	amd_iommu_flush_all_domains();
-- 
1.6.5.3

[PATCH 34/34] x86/amd-iommu: Remove amd_iommu_pd_table

[Joerg Roedel]

The data that was stored in this table is now available in
dev->archdata.iommu. So this table is not longer necessary.
This patch removes the remaining uses of that variable and
removes it from the code.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/include/asm/amd_iommu_types.h |    3 --
 arch/x86/kernel/amd_iommu.c            |   35 ++++++++++---------------------
 arch/x86/kernel/amd_iommu_init.c       |   18 ----------------
 3 files changed, 11 insertions(+), 45 deletions(-)

diff --git a/arch/x86/include/asm/amd_iommu_types.h b/arch/x86/include/asm/amd_iommu_types.h
index f92d1b3..ba19ad4 100644
--- a/arch/x86/include/asm/amd_iommu_types.h
+++ b/arch/x86/include/asm/amd_iommu_types.h
@@ -457,9 +457,6 @@ extern unsigned amd_iommu_aperture_order;
 /* largest PCI device id we expect translation requests for */
 extern u16 amd_iommu_last_bdf;
 
-/* data structures for protection domain handling */
-extern struct protection_domain **amd_iommu_pd_table;
-
 /* allocation bitmap for domain ids */
 extern unsigned long *amd_iommu_pd_alloc_bitmap;
 
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index b75fcd9..32fb091 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -1309,8 +1309,6 @@ static void set_dte_entry(u16 devid, struct protection_domain *domain)
 {
 	u64 pte_root = virt_to_phys(domain->pt_root);
 
-	BUG_ON(amd_iommu_pd_table[devid] != NULL);
-
 	pte_root |= (domain->mode & DEV_ENTRY_MODE_MASK)
 		    << DEV_ENTRY_MODE_SHIFT;
 	pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | IOMMU_PTE_TV;
@@ -1318,20 +1316,10 @@ static void set_dte_entry(u16 devid, struct protection_domain *domain)
 	amd_iommu_dev_table[devid].data[2] = domain->id;
 	amd_iommu_dev_table[devid].data[1] = upper_32_bits(pte_root);
 	amd_iommu_dev_table[devid].data[0] = lower_32_bits(pte_root);
-
-	amd_iommu_pd_table[devid] = domain;
-
 }
 
 static void clear_dte_entry(u16 devid)
 {
-	struct protection_domain *domain = amd_iommu_pd_table[devid];
-
-	BUG_ON(domain == NULL);
-
-	/* remove domain from the lookup table */
-	amd_iommu_pd_table[devid] = NULL;
-
 	/* remove entry from the device table seen by the hardware */
 	amd_iommu_dev_table[devid].data[0] = IOMMU_PTE_P | IOMMU_PTE_TV;
 	amd_iommu_dev_table[devid].data[1] = 0;
@@ -1641,15 +1629,11 @@ static struct protection_domain *get_domain(struct device *dev)
 
 static void update_device_table(struct protection_domain *domain)
 {
-	unsigned long flags;
-	int i;
+	struct iommu_dev_data *dev_data;
 
-	for (i = 0; i <= amd_iommu_last_bdf; ++i) {
-		if (amd_iommu_pd_table[i] != domain)
-			continue;
-		write_lock_irqsave(&amd_iommu_devtable_lock, flags);
-		set_dte_entry(i, domain);
-		write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
+	list_for_each_entry(dev_data, &domain->dev_list, list) {
+		u16 devid = get_device_id(dev_data->dev);
+		set_dte_entry(devid, domain);
 	}
 }
 
@@ -2259,14 +2243,17 @@ free_domains:
 
 static void cleanup_domain(struct protection_domain *domain)
 {
+	struct iommu_dev_data *dev_data, *next;
 	unsigned long flags;
-	u16 devid;
 
 	write_lock_irqsave(&amd_iommu_devtable_lock, flags);
 
-	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
-		if (amd_iommu_pd_table[devid] == domain)
-			clear_dte_entry(devid);
+	list_for_each_entry_safe(dev_data, next, &domain->dev_list, list) {
+		struct device *dev = dev_data->dev;
+
+		do_detach(dev);
+		atomic_set(&dev_data->bind, 0);
+	}
 
 	write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
 }
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index fe1686f..7ffc399 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -165,12 +165,6 @@ u16 *amd_iommu_alias_table;
 struct amd_iommu **amd_iommu_rlookup_table;
 
 /*
- * The pd table (protection domain table) is used to find the protection domain
- * data structure a device belongs to. Indexed with the PCI device id too.
- */
-struct protection_domain **amd_iommu_pd_table;
-
-/*
  * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
  * to know which ones are already in use.
  */
@@ -1238,15 +1232,6 @@ static int __init amd_iommu_init(void)
 	if (amd_iommu_rlookup_table == NULL)
 		goto free;
 
-	/*
-	 * Protection Domain table - maps devices to protection domains
-	 * This table has the same size as the rlookup_table
-	 */
-	amd_iommu_pd_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
-				     get_order(rlookup_table_size));
-	if (amd_iommu_pd_table == NULL)
-		goto free;
-
 	amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
 					    GFP_KERNEL | __GFP_ZERO,
 					    get_order(MAX_DOMAIN_ID/8));
@@ -1314,9 +1299,6 @@ free:
 	free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
 		   get_order(MAX_DOMAIN_ID/8));
 
-	free_pages((unsigned long)amd_iommu_pd_table,
-		   get_order(rlookup_table_size));
-
 	free_pages((unsigned long)amd_iommu_rlookup_table,
 		   get_order(rlookup_table_size));
 
-- 
1.6.5.3

Re: [git pull] IOMMU updates for 2.6.33

[Ingo Molnar]

* Joerg Roedel <joerg.roedel@amd.com> wrote:

> Hi Ingo,
> 
> 
> The following changes since commit 273bee27fa9f79d94b78c83506016f2e41e78983:
>   FUJITA Tomonori (1):
>         x86: Fix iommu=soft boot option
> 
> are available in the git repository at:
> 
>   git://git.kernel.org/pub/scm/linux/kernel/git/joro/linux-2.6-iommu.git amd-iommu/2.6.33
> 
> Joerg Roedel (35):
>       x86/amd-iommu: Separate internal interface definitions
>       x86/amd-iommu: Update copyright headers
>       x86/amd-iommu: Add an index field to struct amd_iommu
>       x86/amd-iommu: Add per IOMMU reference counting
>       x86/amd-iommu: Add function to complete a tlb flush
>       x86/amd-iommu: Make iommu_flush_pages aware of multiple IOMMUs
>       x86/amd-iommu: Use __iommu_flush_pages for tlb flushes
>       x86/amd-iommu: Remove iommu_flush_domain function
>       x86/amd-iommu: Implement protection domain list
>       x86/amd-iommu: Reimplement amd_iommu_flush_all_domains()
>       x86/amd-iommu: Reimplement flush_all_domains_on_iommu()
>       x86/amd-iommu: Make np-cache a global flag
>       x86/amd-iommu: Use check_device for amd_iommu_dma_supported
>       x86/amd-iommu: Use check_device in get_device_resources
>       x86/amd-iommu: Remove iommu parameter from dma_ops_domain_(un)map
>       x86/amd-iommu: Make alloc_new_range aware of multiple IOMMUs
>       x86/amd-iommu: Remove iommu parameter from __(un)map_single
>       x86/amd-iommu: Remove iommu specific handling from dma_ops path
>       x86/amd-iommu: Let domain_for_device handle aliases
>       x86/amd-iommu: Simplify get_device_resources()
>       x86/amd-iommu: Move find_protection_domain to helper functions
>       x86/amd-iommu: Use get_device_id and check_device where appropriate
>       x86/amd-iommu: Remove iommu parameter from dma_ops_domain_alloc
>       x86/amd-iommu: Move some pte allocation functions in the right section
>       x86/amd-iommu: Rearrange dma_ops related functions
>       x86/amd-iommu: Remove support for domain sharing
>       x86/amd-iommu: Use dev->arch->iommu to store iommu related information
>       x86/amd-iommu: Add device bind reference counting
>       x86/amd-iommu: Keep devices per domain in a list
>       x86/amd-iommu: Cleanup attach/detach_device code
>       x86/amd-iommu: Introduce iommu_flush_device() function
>       x86/amd-iommu: Cleanup DTE flushing code
>       x86/amd-iommu: Move reset_iommu_command_buffer out of locked code
>       x86/amd-iommu: Remove amd_iommu_pd_table
>       Merge branch 'gart/fixes' into amd-iommu/2.6.33
> 
> Pavel Vasilyev (1):
>       agp/amd64: Remove GART dependency on AGP_AMD64
> 
>  arch/x86/include/asm/amd_iommu.h       |   12 +-
>  arch/x86/include/asm/amd_iommu_proto.h |   38 +
>  arch/x86/include/asm/amd_iommu_types.h |   54 ++-
>  arch/x86/include/asm/device.h          |    2 +-
>  arch/x86/kernel/amd_iommu.c            | 1230 +++++++++++++++++---------------
>  arch/x86/kernel/amd_iommu_init.c       |   67 +-
>  drivers/char/agp/Kconfig               |    3 +-
>  7 files changed, 777 insertions(+), 629 deletions(-)
>  create mode 100644 arch/x86/include/asm/amd_iommu_proto.h
> 
> Most of the changes are for AMD IOMMU driver. The changes there are 
> bigger than usual because I did a lot of changes to the data 
> structures storing the relations between devices, protection domains 
> and iommus. A major change is that the amd_iommu_pd_table is now gone 
> and replaced by a struct iommu_dev_data for each device stored in 
> dev->archdata->iommu. These changes result in a more robust reference 
> counting, especially for devices with an alias and simplified code in 
> many places. Also included are some related cleanups which separated 
> the internal function prototypes to a several header and code movement 
> to get rid of most function forward declarations. Please pull.

Pulled into tip:core/iommu - thanks a lot Joerg!

	Ingo