[PATCH 0/2 V4] perf/x86/amd: IOMMU Performance Counter Support

[PATCH 0/2 V4] perf/x86/amd: IOMMU Performance Counter Support

[suravee.suthikulpanit]

From: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>

These patches implement the AMD IOMMU Performance Counter functionality
via custom perf PMU and implement static counting for various IOMMU
translations.

    1) Extend the AMD IOMMU initialization to include performance
       counter enablement.

    2) The perf AMD IOMMU PMU to manage performance counters, which
       interface with the AMD IOMMU core driver.

Steven L Kinney (1):
  perf/x86/amd: Adding IOMMU PC resource management

Suravee Suthikulpanit (1):
  perf/x86/amd: AMD IOMMU PC PERF uncore PMU implementation

 arch/x86/kernel/cpu/Makefile               |    4 +
 arch/x86/kernel/cpu/perf_event_amd_iommu.c |  500 ++++++++++++++++++++++++++++
 arch/x86/kernel/cpu/perf_event_amd_iommu.h |   40 +++
 drivers/iommu/amd_iommu_init.c             |  140 +++++++-
 drivers/iommu/amd_iommu_proto.h            |    7 +
 drivers/iommu/amd_iommu_types.h            |   15 +-
 6 files changed, 694 insertions(+), 12 deletions(-)
 create mode 100644 arch/x86/kernel/cpu/perf_event_amd_iommu.c
 create mode 100644 arch/x86/kernel/cpu/perf_event_amd_iommu.h

-- 
1.7.10.4

[PATCH 0/2 V4] perf/x86/amd: IOMMU Performance Counter Support

[suravee.suthikulpanit-5C7GfCeVMHo]

From: Suravee Suthikulpanit <suravee.suthikulpanit-5C7GfCeVMHo@public.gmane.org>

These patches implement the AMD IOMMU Performance Counter functionality
via custom perf PMU and implement static counting for various IOMMU
translations.

    1) Extend the AMD IOMMU initialization to include performance
       counter enablement.

    2) The perf AMD IOMMU PMU to manage performance counters, which
       interface with the AMD IOMMU core driver.

Steven L Kinney (1):
  perf/x86/amd: Adding IOMMU PC resource management

Suravee Suthikulpanit (1):
  perf/x86/amd: AMD IOMMU PC PERF uncore PMU implementation

 arch/x86/kernel/cpu/Makefile               |    4 +
 arch/x86/kernel/cpu/perf_event_amd_iommu.c |  500 ++++++++++++++++++++++++++++
 arch/x86/kernel/cpu/perf_event_amd_iommu.h |   40 +++
 drivers/iommu/amd_iommu_init.c             |  140 +++++++-
 drivers/iommu/amd_iommu_proto.h            |    7 +
 drivers/iommu/amd_iommu_types.h            |   15 +-
 6 files changed, 694 insertions(+), 12 deletions(-)
 create mode 100644 arch/x86/kernel/cpu/perf_event_amd_iommu.c
 create mode 100644 arch/x86/kernel/cpu/perf_event_amd_iommu.h

-- 
1.7.10.4

[PATCH 1/2 V4] perf/x86/amd: Adding IOMMU PC resource management

[suravee.suthikulpanit]

From: Steven L Kinney <Steven.Kinney@amd.com>

Add functionality to check the availability of the AMD IOMMU Performance
Counters and export this functionality to other core drivers, such as in this
case, a perf AMD IOMMU PMU.  This feature is not bound to any specific AMD
family/model other than the presence of the IOMMU with PC enabled.

The AMD IOMMU PC support static counting only at this time.

Signed-off-by: Steven Kinney <steven.kinney@amd.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
---
Changes from V3:
	- Add logic to check EFR for max banks/counters.
	- Address misceleneous changes from Joerg comments.

 drivers/iommu/amd_iommu_init.c  |  140 ++++++++++++++++++++++++++++++++++++---
 drivers/iommu/amd_iommu_proto.h |    7 ++
 drivers/iommu/amd_iommu_types.h |   15 ++++-
 3 files changed, 150 insertions(+), 12 deletions(-)

diff --git a/drivers/iommu/amd_iommu_init.c b/drivers/iommu/amd_iommu_init.c
index bf51abb..7acbc391 100644
--- a/drivers/iommu/amd_iommu_init.c
+++ b/drivers/iommu/amd_iommu_init.c
@@ -99,7 +99,7 @@ struct ivhd_header {
 	u64 mmio_phys;
 	u16 pci_seg;
 	u16 info;
-	u32 reserved;
+	u32 efr;
 } __attribute__((packed));
 
 /*
@@ -154,6 +154,7 @@ bool amd_iommu_iotlb_sup __read_mostly = true;
 u32 amd_iommu_max_pasids __read_mostly = ~0;
 
 bool amd_iommu_v2_present __read_mostly;
+bool amd_iommu_pc_present __read_mostly;
 
 bool amd_iommu_force_isolation __read_mostly;
 
@@ -369,23 +370,23 @@ static void iommu_disable(struct amd_iommu *iommu)
  * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
  * the system has one.
  */
-static u8 __iomem * __init iommu_map_mmio_space(u64 address)
+static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
 {
-	if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu")) {
-		pr_err("AMD-Vi: Can not reserve memory region %llx for mmio\n",
-			address);
+	if (!request_mem_region(address, end, "amd_iommu")) {
+		pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
+			address, end);
 		pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
 		return NULL;
 	}
 
-	return (u8 __iomem *)ioremap_nocache(address, MMIO_REGION_LENGTH);
+	return (u8 __iomem *)ioremap_nocache(address, end);
 }
 
 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
 {
 	if (iommu->mmio_base)
 		iounmap(iommu->mmio_base);
-	release_mem_region(iommu->mmio_phys, MMIO_REGION_LENGTH);
+	release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
 }
 
 /****************************************************************************
@@ -1085,7 +1086,18 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
 	iommu->cap_ptr = h->cap_ptr;
 	iommu->pci_seg = h->pci_seg;
 	iommu->mmio_phys = h->mmio_phys;
-	iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
+
+	/* Check if IVHD EFR contains proper max banks/counters */
+	if ((h->efr != 0) &&
+	    ((h->efr & (0xF << 13)) != 0) &&
+	    ((h->efr & (0x3F << 17)) != 0)) {
+		iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
+	} else {
+		iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
+	}
+
+	iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
+						iommu->mmio_phys_end);
 	if (!iommu->mmio_base)
 		return -ENOMEM;
 
@@ -1160,6 +1172,33 @@ static int __init init_iommu_all(struct acpi_table_header *table)
 	return 0;
 }
 
+
+static void init_iommu_perf_ctr(struct amd_iommu *iommu)
+{
+	u64 val = 0xabcd, val2 = 0;
+
+	if (!iommu_feature(iommu, FEATURE_PC))
+		return;
+
+	amd_iommu_pc_present = true;
+
+	/* Check if the performance counters can be written to */
+	if ((0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val, true)) ||
+	    (0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val2, false)) ||
+	    (val != val2)) {
+		pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
+		amd_iommu_pc_present = false;
+		return;
+	}
+
+	pr_info("AMD-Vi: IOMMU performance counters supported\n");
+
+	val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
+	iommu->max_banks = (u8) ((val >> 12) & 0x3f);
+	iommu->max_counters = (u8) ((val >> 7) & 0xf);
+}
+
+
 static int iommu_init_pci(struct amd_iommu *iommu)
 {
 	int cap_ptr = iommu->cap_ptr;
@@ -1226,6 +1265,8 @@ static int iommu_init_pci(struct amd_iommu *iommu)
 	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
 		amd_iommu_np_cache = true;
 
+	init_iommu_perf_ctr(iommu);
+
 	if (is_rd890_iommu(iommu->dev)) {
 		int i, j;
 
@@ -1278,7 +1319,7 @@ static void print_iommu_info(void)
 				if (iommu_feature(iommu, (1ULL << i)))
 					pr_cont(" %s", feat_str[i]);
 			}
-		pr_cont("\n");
+			pr_cont("\n");
 		}
 	}
 	if (irq_remapping_enabled)
@@ -2232,3 +2273,84 @@ bool amd_iommu_v2_supported(void)
 	return amd_iommu_v2_present;
 }
 EXPORT_SYMBOL(amd_iommu_v2_supported);
+
+/****************************************************************************
+ *
+ * IOMMU EFR Performance Counter support functionality. This code allows
+ * access to the IOMMU PC functionality.
+ *
+ ****************************************************************************/
+
+u8 amd_iommu_pc_get_max_banks(u16 devid)
+{
+	struct amd_iommu *iommu;
+	u8 ret = 0;
+
+	/* locate the iommu governing the devid */
+	iommu = amd_iommu_rlookup_table[devid];
+	if (iommu)
+		ret = iommu->max_banks;
+
+	return ret;
+}
+EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
+
+bool amd_iommu_pc_supported(void)
+{
+	return amd_iommu_pc_present;
+}
+EXPORT_SYMBOL(amd_iommu_pc_supported);
+
+u8 amd_iommu_pc_get_max_counters(u16 devid)
+{
+	struct amd_iommu *iommu;
+	u8 ret = 0;
+
+	/* locate the iommu governing the devid */
+	iommu = amd_iommu_rlookup_table[devid];
+	if (iommu)
+		ret = iommu->max_counters;
+
+	return ret;
+}
+EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
+
+int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr, u8 fxn,
+				    u64 *value, bool is_write)
+{
+	struct amd_iommu *iommu;
+	u32 offset;
+	u32 max_offset_lim;
+
+	/* Make sure the IOMMU PC resource is available */
+	if (!amd_iommu_pc_present)
+		return -ENODEV;
+
+	/* Locate the iommu associated with the device ID */
+	iommu = amd_iommu_rlookup_table[devid];
+
+	/* Check for valid iommu and pc register indexing */
+	if (WARN_ON((iommu == NULL) || (fxn > 0x28) || (fxn & 7)))
+		return -ENODEV;
+
+	offset = (u32)(((0x40|bank) << 12) | (cntr << 8) | fxn);
+
+	/* Limit the offset to the hw defined mmio region aperture */
+	max_offset_lim = (u32)(((0x40|iommu->max_banks) << 12) |
+				(iommu->max_counters << 8) | 0x28);
+	if ((offset < MMIO_CNTR_REG_OFFSET) ||
+	    (offset > max_offset_lim))
+		return -EINVAL;
+
+	if (is_write) {
+		writel((u32)*value, iommu->mmio_base + offset);
+		writel((*value >> 32), iommu->mmio_base + offset + 4);
+	} else {
+		*value = readl(iommu->mmio_base + offset + 4);
+		*value <<= 32;
+		*value = readl(iommu->mmio_base + offset);
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL(amd_iommu_pc_get_set_reg_val);
diff --git a/drivers/iommu/amd_iommu_proto.h b/drivers/iommu/amd_iommu_proto.h
index c294961..95ed6de 100644
--- a/drivers/iommu/amd_iommu_proto.h
+++ b/drivers/iommu/amd_iommu_proto.h
@@ -56,6 +56,13 @@ extern int amd_iommu_domain_set_gcr3(struct iommu_domain *dom, int pasid,
 extern int amd_iommu_domain_clear_gcr3(struct iommu_domain *dom, int pasid);
 extern struct iommu_domain *amd_iommu_get_v2_domain(struct pci_dev *pdev);
 
+/* IOMMU Performance Counter functions */
+extern bool amd_iommu_pc_supported(void);
+extern u8 amd_iommu_pc_get_max_banks(u16 devid);
+extern u8 amd_iommu_pc_get_max_counters(u16 devid);
+extern int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr, u8 fxn,
+				    u64 *value, bool is_write);
+
 #define PPR_SUCCESS			0x0
 #define PPR_INVALID			0x1
 #define PPR_FAILURE			0xf
diff --git a/drivers/iommu/amd_iommu_types.h b/drivers/iommu/amd_iommu_types.h
index 0285a21..d901832 100644
--- a/drivers/iommu/amd_iommu_types.h
+++ b/drivers/iommu/amd_iommu_types.h
@@ -38,9 +38,6 @@
 #define ALIAS_TABLE_ENTRY_SIZE		2
 #define RLOOKUP_TABLE_ENTRY_SIZE	(sizeof(void *))
 
-/* Length of the MMIO region for the AMD IOMMU */
-#define MMIO_REGION_LENGTH       0x4000
-
 /* Capability offsets used by the driver */
 #define MMIO_CAP_HDR_OFFSET	0x00
 #define MMIO_RANGE_OFFSET	0x0c
@@ -78,6 +75,10 @@
 #define MMIO_STATUS_OFFSET	0x2020
 #define MMIO_PPR_HEAD_OFFSET	0x2030
 #define MMIO_PPR_TAIL_OFFSET	0x2038
+#define MMIO_CNTR_CONF_OFFSET	0x4000
+#define MMIO_CNTR_REG_OFFSET	0x40000
+#define MMIO_REG_END_OFFSET	0x80000
+
 
 
 /* Extended Feature Bits */
@@ -507,6 +508,10 @@ struct amd_iommu {
 
 	/* physical address of MMIO space */
 	u64 mmio_phys;
+
+	/* physical end address of MMIO space */
+	u64 mmio_phys_end;
+
 	/* virtual address of MMIO space */
 	u8 __iomem *mmio_base;
 
@@ -584,6 +589,10 @@ struct amd_iommu {
 
 	/* The l2 indirect registers */
 	u32 stored_l2[0x83];
+
+	/* The maximum PC banks and counters/bank (PCSup=1) */
+	u8 max_banks;
+	u8 max_counters;
 };
 
 struct devid_map {
-- 
1.7.10.4

[PATCH 1/2 V4] perf/x86/amd: Adding IOMMU PC resource management

[suravee.suthikulpanit-5C7GfCeVMHo]

From: Steven L Kinney <Steven.Kinney-5C7GfCeVMHo@public.gmane.org>

Add functionality to check the availability of the AMD IOMMU Performance
Counters and export this functionality to other core drivers, such as in this
case, a perf AMD IOMMU PMU.  This feature is not bound to any specific AMD
family/model other than the presence of the IOMMU with PC enabled.

The AMD IOMMU PC support static counting only at this time.

Signed-off-by: Steven Kinney <steven.kinney-5C7GfCeVMHo@public.gmane.org>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit-5C7GfCeVMHo@public.gmane.org>
---
Changes from V3:
	- Add logic to check EFR for max banks/counters.
	- Address misceleneous changes from Joerg comments.

 drivers/iommu/amd_iommu_init.c  |  140 ++++++++++++++++++++++++++++++++++++---
 drivers/iommu/amd_iommu_proto.h |    7 ++
 drivers/iommu/amd_iommu_types.h |   15 ++++-
 3 files changed, 150 insertions(+), 12 deletions(-)

diff --git a/drivers/iommu/amd_iommu_init.c b/drivers/iommu/amd_iommu_init.c
index bf51abb..7acbc391 100644
--- a/drivers/iommu/amd_iommu_init.c
+++ b/drivers/iommu/amd_iommu_init.c
@@ -99,7 +99,7 @@ struct ivhd_header {
 	u64 mmio_phys;
 	u16 pci_seg;
 	u16 info;
-	u32 reserved;
+	u32 efr;
 } __attribute__((packed));
 
 /*
@@ -154,6 +154,7 @@ bool amd_iommu_iotlb_sup __read_mostly = true;
 u32 amd_iommu_max_pasids __read_mostly = ~0;
 
 bool amd_iommu_v2_present __read_mostly;
+bool amd_iommu_pc_present __read_mostly;
 
 bool amd_iommu_force_isolation __read_mostly;
 
@@ -369,23 +370,23 @@ static void iommu_disable(struct amd_iommu *iommu)
  * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
  * the system has one.
  */
-static u8 __iomem * __init iommu_map_mmio_space(u64 address)
+static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
 {
-	if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu")) {
-		pr_err("AMD-Vi: Can not reserve memory region %llx for mmio\n",
-			address);
+	if (!request_mem_region(address, end, "amd_iommu")) {
+		pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
+			address, end);
 		pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
 		return NULL;
 	}
 
-	return (u8 __iomem *)ioremap_nocache(address, MMIO_REGION_LENGTH);
+	return (u8 __iomem *)ioremap_nocache(address, end);
 }
 
 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
 {
 	if (iommu->mmio_base)
 		iounmap(iommu->mmio_base);
-	release_mem_region(iommu->mmio_phys, MMIO_REGION_LENGTH);
+	release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
 }
 
 /****************************************************************************
@@ -1085,7 +1086,18 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
 	iommu->cap_ptr = h->cap_ptr;
 	iommu->pci_seg = h->pci_seg;
 	iommu->mmio_phys = h->mmio_phys;
-	iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
+
+	/* Check if IVHD EFR contains proper max banks/counters */
+	if ((h->efr != 0) &&
+	    ((h->efr & (0xF << 13)) != 0) &&
+	    ((h->efr & (0x3F << 17)) != 0)) {
+		iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
+	} else {
+		iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
+	}
+
+	iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
+						iommu->mmio_phys_end);
 	if (!iommu->mmio_base)
 		return -ENOMEM;
 
@@ -1160,6 +1172,33 @@ static int __init init_iommu_all(struct acpi_table_header *table)
 	return 0;
 }
 
+
+static void init_iommu_perf_ctr(struct amd_iommu *iommu)
+{
+	u64 val = 0xabcd, val2 = 0;
+
+	if (!iommu_feature(iommu, FEATURE_PC))
+		return;
+
+	amd_iommu_pc_present = true;
+
+	/* Check if the performance counters can be written to */
+	if ((0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val, true)) ||
+	    (0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val2, false)) ||
+	    (val != val2)) {
+		pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
+		amd_iommu_pc_present = false;
+		return;
+	}
+
+	pr_info("AMD-Vi: IOMMU performance counters supported\n");
+
+	val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
+	iommu->max_banks = (u8) ((val >> 12) & 0x3f);
+	iommu->max_counters = (u8) ((val >> 7) & 0xf);
+}
+
+
 static int iommu_init_pci(struct amd_iommu *iommu)
 {
 	int cap_ptr = iommu->cap_ptr;
@@ -1226,6 +1265,8 @@ static int iommu_init_pci(struct amd_iommu *iommu)
 	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
 		amd_iommu_np_cache = true;
 
+	init_iommu_perf_ctr(iommu);
+
 	if (is_rd890_iommu(iommu->dev)) {
 		int i, j;
 
@@ -1278,7 +1319,7 @@ static void print_iommu_info(void)
 				if (iommu_feature(iommu, (1ULL << i)))
 					pr_cont(" %s", feat_str[i]);
 			}
-		pr_cont("\n");
+			pr_cont("\n");
 		}
 	}
 	if (irq_remapping_enabled)
@@ -2232,3 +2273,84 @@ bool amd_iommu_v2_supported(void)
 	return amd_iommu_v2_present;
 }
 EXPORT_SYMBOL(amd_iommu_v2_supported);
+
+/****************************************************************************
+ *
+ * IOMMU EFR Performance Counter support functionality. This code allows
+ * access to the IOMMU PC functionality.
+ *
+ ****************************************************************************/
+
+u8 amd_iommu_pc_get_max_banks(u16 devid)
+{
+	struct amd_iommu *iommu;
+	u8 ret = 0;
+
+	/* locate the iommu governing the devid */
+	iommu = amd_iommu_rlookup_table[devid];
+	if (iommu)
+		ret = iommu->max_banks;
+
+	return ret;
+}
+EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
+
+bool amd_iommu_pc_supported(void)
+{
+	return amd_iommu_pc_present;
+}
+EXPORT_SYMBOL(amd_iommu_pc_supported);
+
+u8 amd_iommu_pc_get_max_counters(u16 devid)
+{
+	struct amd_iommu *iommu;
+	u8 ret = 0;
+
+	/* locate the iommu governing the devid */
+	iommu = amd_iommu_rlookup_table[devid];
+	if (iommu)
+		ret = iommu->max_counters;
+
+	return ret;
+}
+EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
+
+int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr, u8 fxn,
+				    u64 *value, bool is_write)
+{
+	struct amd_iommu *iommu;
+	u32 offset;
+	u32 max_offset_lim;
+
+	/* Make sure the IOMMU PC resource is available */
+	if (!amd_iommu_pc_present)
+		return -ENODEV;
+
+	/* Locate the iommu associated with the device ID */
+	iommu = amd_iommu_rlookup_table[devid];
+
+	/* Check for valid iommu and pc register indexing */
+	if (WARN_ON((iommu == NULL) || (fxn > 0x28) || (fxn & 7)))
+		return -ENODEV;
+
+	offset = (u32)(((0x40|bank) << 12) | (cntr << 8) | fxn);
+
+	/* Limit the offset to the hw defined mmio region aperture */
+	max_offset_lim = (u32)(((0x40|iommu->max_banks) << 12) |
+				(iommu->max_counters << 8) | 0x28);
+	if ((offset < MMIO_CNTR_REG_OFFSET) ||
+	    (offset > max_offset_lim))
+		return -EINVAL;
+
+	if (is_write) {
+		writel((u32)*value, iommu->mmio_base + offset);
+		writel((*value >> 32), iommu->mmio_base + offset + 4);
+	} else {
+		*value = readl(iommu->mmio_base + offset + 4);
+		*value <<= 32;
+		*value = readl(iommu->mmio_base + offset);
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL(amd_iommu_pc_get_set_reg_val);
diff --git a/drivers/iommu/amd_iommu_proto.h b/drivers/iommu/amd_iommu_proto.h
index c294961..95ed6de 100644
--- a/drivers/iommu/amd_iommu_proto.h
+++ b/drivers/iommu/amd_iommu_proto.h
@@ -56,6 +56,13 @@ extern int amd_iommu_domain_set_gcr3(struct iommu_domain *dom, int pasid,
 extern int amd_iommu_domain_clear_gcr3(struct iommu_domain *dom, int pasid);
 extern struct iommu_domain *amd_iommu_get_v2_domain(struct pci_dev *pdev);
 
+/* IOMMU Performance Counter functions */
+extern bool amd_iommu_pc_supported(void);
+extern u8 amd_iommu_pc_get_max_banks(u16 devid);
+extern u8 amd_iommu_pc_get_max_counters(u16 devid);
+extern int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr, u8 fxn,
+				    u64 *value, bool is_write);
+
 #define PPR_SUCCESS			0x0
 #define PPR_INVALID			0x1
 #define PPR_FAILURE			0xf
diff --git a/drivers/iommu/amd_iommu_types.h b/drivers/iommu/amd_iommu_types.h
index 0285a21..d901832 100644
--- a/drivers/iommu/amd_iommu_types.h
+++ b/drivers/iommu/amd_iommu_types.h
@@ -38,9 +38,6 @@
 #define ALIAS_TABLE_ENTRY_SIZE		2
 #define RLOOKUP_TABLE_ENTRY_SIZE	(sizeof(void *))
 
-/* Length of the MMIO region for the AMD IOMMU */
-#define MMIO_REGION_LENGTH       0x4000
-
 /* Capability offsets used by the driver */
 #define MMIO_CAP_HDR_OFFSET	0x00
 #define MMIO_RANGE_OFFSET	0x0c
@@ -78,6 +75,10 @@
 #define MMIO_STATUS_OFFSET	0x2020
 #define MMIO_PPR_HEAD_OFFSET	0x2030
 #define MMIO_PPR_TAIL_OFFSET	0x2038
+#define MMIO_CNTR_CONF_OFFSET	0x4000
+#define MMIO_CNTR_REG_OFFSET	0x40000
+#define MMIO_REG_END_OFFSET	0x80000
+
 
 
 /* Extended Feature Bits */
@@ -507,6 +508,10 @@ struct amd_iommu {
 
 	/* physical address of MMIO space */
 	u64 mmio_phys;
+
+	/* physical end address of MMIO space */
+	u64 mmio_phys_end;
+
 	/* virtual address of MMIO space */
 	u8 __iomem *mmio_base;
 
@@ -584,6 +589,10 @@ struct amd_iommu {
 
 	/* The l2 indirect registers */
 	u32 stored_l2[0x83];
+
+	/* The maximum PC banks and counters/bank (PCSup=1) */
+	u8 max_banks;
+	u8 max_counters;
 };
 
 struct devid_map {
-- 
1.7.10.4

[PATCH 2/2 V4] perf/x86/amd: AMD IOMMU PC PERF uncore PMU implementation

[suravee.suthikulpanit]

From: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>

Implement a perf PMU to handle IOMMU performance counters and events.
The PMU only supports counting mode (e.g. perf stat). Since the counters
are shared across all cores, the PMU is implemented as "system-wide" mode.

To invoke the AMD IOMMU PMU, issue a perf tool command such as:

./perf stat -a -e amd_iommu/<events>/ <command>
or
./perf stat -a -e amd_iommu/config=<config-data>,config1=<config1-data>/ <command>

For example:

./perf stat -a -e amd_iommu/mem_trans_total/ <command>

The resulting count will be how many IOMMU total peripheral memory
operations were performed during the command execution window.

Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
---
Changes from V3:
	- Add CONFIG_AMD_IOMMU in the Makefile
	- Add proper check for max_banks and max_counters

 arch/x86/kernel/cpu/Makefile               |    4 +
 arch/x86/kernel/cpu/perf_event_amd_iommu.c |  500 ++++++++++++++++++++++++++++
 arch/x86/kernel/cpu/perf_event_amd_iommu.h |   40 +++
 3 files changed, 544 insertions(+)
 create mode 100644 arch/x86/kernel/cpu/perf_event_amd_iommu.c
 create mode 100644 arch/x86/kernel/cpu/perf_event_amd_iommu.h

diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index b0684e4..47b56a7 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -31,11 +31,15 @@ obj-$(CONFIG_PERF_EVENTS)		+= perf_event.o
 
 ifdef CONFIG_PERF_EVENTS
 obj-$(CONFIG_CPU_SUP_AMD)		+= perf_event_amd.o perf_event_amd_uncore.o
+ifdef CONFIG_AMD_IOMMU
+obj-$(CONFIG_CPU_SUP_AMD)		+= perf_event_amd_iommu.o
+endif
 obj-$(CONFIG_CPU_SUP_INTEL)		+= perf_event_p6.o perf_event_knc.o perf_event_p4.o
 obj-$(CONFIG_CPU_SUP_INTEL)		+= perf_event_intel_lbr.o perf_event_intel_ds.o perf_event_intel.o
 obj-$(CONFIG_CPU_SUP_INTEL)		+= perf_event_intel_uncore.o
 endif
 
+
 obj-$(CONFIG_X86_MCE)			+= mcheck/
 obj-$(CONFIG_MTRR)			+= mtrr/
 
diff --git a/arch/x86/kernel/cpu/perf_event_amd_iommu.c b/arch/x86/kernel/cpu/perf_event_amd_iommu.c
new file mode 100644
index 0000000..a1199cd
--- /dev/null
+++ b/arch/x86/kernel/cpu/perf_event_amd_iommu.c
@@ -0,0 +1,500 @@
+/*
+ * Copyright (C) 2013 Advanced Micro Devices, Inc.
+ *
+ * Author: Steven Kinney <Steven.Kinney@amd.com>
+ * Author: Suravee Suthikulpanit <Suraveee.Suthikulpanit@amd.com>
+ *
+ * Perf: amd_iommu - AMD IOMMU Performance Counter PMU implementation
+ *
+ * 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.
+ */
+
+#include <linux/perf_event.h>
+#include <linux/module.h>
+#include <linux/cpumask.h>
+#include <linux/slab.h>
+
+#include "perf_event.h"
+#include "perf_event_amd_iommu.h"
+
+#define _GET_BANK(ev)       ((u8)(ev->hw.extra_reg.reg >> 8))
+#define _GET_CNTR(ev)       ((u8)(ev->hw.extra_reg.reg))
+
+/* iommu pmu config masks */
+#define _GET_CSOURCE(ev)    ((ev->hw.config & 0xFFULL))
+#define _GET_DEVID(ev)      ((ev->hw.config >> 8)  & 0xFFFFULL)
+#define _GET_PASID(ev)      ((ev->hw.config >> 24) & 0xFFFFULL)
+#define _GET_DOMID(ev)      ((ev->hw.config >> 40) & 0xFFFFULL)
+#define _GET_DEVID_MASK(ev) ((ev->hw.extra_reg.config)  & 0xFFFFULL)
+#define _GET_PASID_MASK(ev) ((ev->hw.extra_reg.config >> 16) & 0xFFFFULL)
+#define _GET_DOMID_MASK(ev) ((ev->hw.extra_reg.config >> 32) & 0xFFFFULL)
+
+static struct perf_amd_iommu __perf_iommu;
+
+struct perf_amd_iommu {
+	struct pmu pmu;
+	u8 max_banks;
+	u8 max_counters;
+	u64 cntr_assign_mask;
+	raw_spinlock_t lock;
+	const struct attribute_group *attr_groups[4];
+};
+
+#define format_group	attr_groups[0]
+#define cpumask_group	attr_groups[1]
+#define events_group	attr_groups[2]
+#define null_group	attr_groups[3]
+
+/*---------------------------------------------
+ * sysfs format attributes
+ *---------------------------------------------*/
+PMU_FORMAT_ATTR(csource,    "config:0-7");
+PMU_FORMAT_ATTR(devid,      "config:8-23");
+PMU_FORMAT_ATTR(pasid,      "config:24-39");
+PMU_FORMAT_ATTR(domid,      "config:40-55");
+PMU_FORMAT_ATTR(devid_mask, "config1:0-15");
+PMU_FORMAT_ATTR(pasid_mask, "config1:16-31");
+PMU_FORMAT_ATTR(domid_mask, "config1:32-47");
+
+static struct attribute *iommu_format_attrs[] = {
+	&format_attr_csource.attr,
+	&format_attr_devid.attr,
+	&format_attr_pasid.attr,
+	&format_attr_domid.attr,
+	&format_attr_devid_mask.attr,
+	&format_attr_pasid_mask.attr,
+	&format_attr_domid_mask.attr,
+	NULL,
+};
+
+static struct attribute_group amd_iommu_format_group = {
+	.name = "format",
+	.attrs = iommu_format_attrs,
+};
+
+/*---------------------------------------------
+ * sysfs events attributes
+ *---------------------------------------------*/
+struct amd_iommu_event_desc {
+	struct kobj_attribute attr;
+	const char *event;
+};
+
+static ssize_t _iommu_event_show(struct kobject *kobj,
+				struct kobj_attribute *attr, char *buf)
+{
+	struct amd_iommu_event_desc *event =
+		container_of(attr, struct amd_iommu_event_desc, attr);
+	return sprintf(buf, "%s\n", event->event);
+}
+
+#define AMD_IOMMU_EVENT_DESC(_name, _event)			\
+{								\
+	.attr  = __ATTR(_name, 0444, _iommu_event_show, NULL),	\
+	.event = _event,					\
+}
+
+static struct amd_iommu_event_desc amd_iommu_v2_event_descs[] = {
+	AMD_IOMMU_EVENT_DESC(mem_pass_untrans,        "csource=0x01"),
+	AMD_IOMMU_EVENT_DESC(mem_pass_pretrans,       "csource=0x02"),
+	AMD_IOMMU_EVENT_DESC(mem_pass_excl,           "csource=0x03"),
+	AMD_IOMMU_EVENT_DESC(mem_target_abort,        "csource=0x04"),
+	AMD_IOMMU_EVENT_DESC(mem_trans_total,         "csource=0x05"),
+	AMD_IOMMU_EVENT_DESC(mem_iommu_tlb_pte_hit,   "csource=0x06"),
+	AMD_IOMMU_EVENT_DESC(mem_iommu_tlb_pte_mis,   "csource=0x07"),
+	AMD_IOMMU_EVENT_DESC(mem_iommu_tlb_pde_hit,   "csource=0x08"),
+	AMD_IOMMU_EVENT_DESC(mem_iommu_tlb_pde_mis,   "csource=0x09"),
+	AMD_IOMMU_EVENT_DESC(mem_dte_hit,             "csource=0x0a"),
+	AMD_IOMMU_EVENT_DESC(mem_dte_mis,             "csource=0x0b"),
+	AMD_IOMMU_EVENT_DESC(page_tbl_read_tot,       "csource=0x0c"),
+	AMD_IOMMU_EVENT_DESC(page_tbl_read_nst,       "csource=0x0d"),
+	AMD_IOMMU_EVENT_DESC(page_tbl_read_gst,       "csource=0x0e"),
+	AMD_IOMMU_EVENT_DESC(int_dte_hit,             "csource=0x0f"),
+	AMD_IOMMU_EVENT_DESC(int_dte_mis,             "csource=0x10"),
+	AMD_IOMMU_EVENT_DESC(cmd_processed,           "csource=0x11"),
+	AMD_IOMMU_EVENT_DESC(cmd_processed_inv,       "csource=0x12"),
+	AMD_IOMMU_EVENT_DESC(tlb_inv,                 "csource=0x13"),
+	{ /* end: all zeroes */ },
+};
+
+/*---------------------------------------------
+ * sysfs cpumask attributes
+ *---------------------------------------------*/
+static cpumask_t iommu_cpumask;
+
+static ssize_t _iommu_cpumask_show(struct device *dev,
+				   struct device_attribute *attr,
+				   char *buf)
+{
+	int n = cpulist_scnprintf(buf, PAGE_SIZE - 2, &iommu_cpumask);
+	buf[n++] = '\n';
+	buf[n] = '\0';
+	return n;
+}
+static DEVICE_ATTR(cpumask, S_IRUGO, _iommu_cpumask_show, NULL);
+
+static struct attribute *iommu_cpumask_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL,
+};
+
+static struct attribute_group amd_iommu_cpumask_group = {
+	.attrs = iommu_cpumask_attrs,
+};
+
+/*---------------------------------------------*/
+
+static int get_next_avail_iommu_bnk_cntr(struct perf_amd_iommu *perf_iommu)
+{
+	unsigned long flags;
+	int shift, bank, cntr, retval;
+	int max_banks = perf_iommu->max_banks;
+	int max_cntrs = perf_iommu->max_counters;
+
+	raw_spin_lock_irqsave(&perf_iommu->lock, flags);
+
+	for (bank = 0, shift = 0; bank < max_banks; bank++) {
+		for (cntr = 0; cntr < max_cntrs; cntr++) {
+			shift = bank + (bank*3) + cntr;
+			if (perf_iommu->cntr_assign_mask & (1ULL<<shift)) {
+				continue;
+			} else {
+				perf_iommu->cntr_assign_mask |= (1ULL<<shift);
+				retval = ((u16)((u16)bank<<8) | (u8)(cntr));
+				goto out;
+			}
+		}
+	}
+	retval = -ENOSPC;
+out:
+	raw_spin_unlock_irqrestore(&perf_iommu->lock, flags);
+	return retval;
+}
+
+static int clear_avail_iommu_bnk_cntr(struct perf_amd_iommu *perf_iommu,
+					u8 bank, u8 cntr)
+{
+	unsigned long flags;
+	int max_banks, max_cntrs;
+	int shift = 0;
+
+	max_banks = perf_iommu->max_banks;
+	max_cntrs = perf_iommu->max_counters;
+
+	if ((bank > max_banks) || (cntr > max_cntrs))
+		return -EINVAL;
+
+	shift = bank + cntr + (bank*3);
+
+	raw_spin_lock_irqsave(&perf_iommu->lock, flags);
+	perf_iommu->cntr_assign_mask &= ~(1ULL<<shift);
+	raw_spin_unlock_irqrestore(&perf_iommu->lock, flags);
+
+	return 0;
+}
+
+static int perf_iommu_event_init(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct perf_amd_iommu *perf_iommu;
+	u64 config, config1;
+
+	/* test the event attr type check for PMU enumeration */
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	/*
+	 * IOMMU counters are shared across all cores.
+	 * Therefore, it does not support per-process mode.
+	 * Also, it does not support event sampling mode.
+	 */
+	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
+		return -EINVAL;
+
+	/* IOMMU counters do not have usr/os/guest/host bits */
+	if (event->attr.exclude_user || event->attr.exclude_kernel ||
+	    event->attr.exclude_host || event->attr.exclude_guest)
+		return -EINVAL;
+
+	if (event->cpu < 0)
+		return -EINVAL;
+
+	perf_iommu = &__perf_iommu;
+
+	if (event->pmu != &perf_iommu->pmu)
+		return -ENOENT;
+
+	if (perf_iommu) {
+		config = event->attr.config;
+		config1 = event->attr.config1;
+	} else {
+		return -EINVAL;
+	}
+
+	/* integrate with iommu base devid (0000), assume one iommu */
+	perf_iommu->max_banks =
+		amd_iommu_pc_get_max_banks(IOMMU_BASE_DEVID);
+	perf_iommu->max_counters =
+		amd_iommu_pc_get_max_counters(IOMMU_BASE_DEVID);
+	if ((perf_iommu->max_banks == 0) || (perf_iommu->max_counters == 0))
+		return -EINVAL;
+
+	/* update the hw_perf_event struct with the iommu config data */
+	hwc->config = config;
+	hwc->extra_reg.config = config1;
+
+	return 0;
+}
+
+static void perf_iommu_enable_event(struct perf_event *ev)
+{
+	u8 csource = _GET_CSOURCE(ev);
+	u16 devid = _GET_DEVID(ev);
+	u64 reg = 0ULL;
+
+	reg = csource;
+	amd_iommu_pc_get_set_reg_val(devid,
+			_GET_BANK(ev), _GET_CNTR(ev) ,
+			 IOMMU_PC_COUNTER_SRC_REG, &reg, true);
+
+	reg = 0ULL | devid | (_GET_DEVID_MASK(ev) << 32);
+	if (reg)
+		reg |= (1UL << 31);
+	amd_iommu_pc_get_set_reg_val(devid,
+			_GET_BANK(ev), _GET_CNTR(ev) ,
+			 IOMMU_PC_DEVID_MATCH_REG, &reg, true);
+
+	reg = 0ULL | _GET_PASID(ev) | (_GET_PASID_MASK(ev) << 32);
+	if (reg)
+		reg |= (1UL << 31);
+	amd_iommu_pc_get_set_reg_val(devid,
+			_GET_BANK(ev), _GET_CNTR(ev) ,
+			 IOMMU_PC_PASID_MATCH_REG, &reg, true);
+
+	reg = 0ULL | _GET_DOMID(ev) | (_GET_DOMID_MASK(ev) << 32);
+	if (reg)
+		reg |= (1UL << 31);
+	amd_iommu_pc_get_set_reg_val(devid,
+			_GET_BANK(ev), _GET_CNTR(ev) ,
+			 IOMMU_PC_DOMID_MATCH_REG, &reg, true);
+}
+
+static void perf_iommu_disable_event(struct perf_event *event)
+{
+	u64 reg = 0ULL;
+
+	amd_iommu_pc_get_set_reg_val(_GET_DEVID(event),
+			_GET_BANK(event), _GET_CNTR(event),
+			IOMMU_PC_COUNTER_SRC_REG, &reg, true);
+}
+
+static void perf_iommu_start(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+
+	pr_debug("perf: amd_iommu:perf_iommu_start\n");
+	if (WARN_ON_ONCE(!(hwc->state & PERF_HES_STOPPED)))
+		return;
+
+	WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));
+	hwc->state = 0;
+
+	if (flags & PERF_EF_RELOAD) {
+		u64 prev_raw_count =  local64_read(&hwc->prev_count);
+		amd_iommu_pc_get_set_reg_val(_GET_DEVID(event),
+				_GET_BANK(event), _GET_CNTR(event),
+				IOMMU_PC_COUNTER_REG, &prev_raw_count, true);
+	}
+
+	perf_iommu_enable_event(event);
+	perf_event_update_userpage(event);
+
+}
+
+static void perf_iommu_read(struct perf_event *event)
+{
+	u64 count = 0ULL;
+	u64 prev_raw_count = 0ULL;
+	u64 delta = 0ULL;
+	struct hw_perf_event *hwc = &event->hw;
+	pr_debug("perf: amd_iommu:perf_iommu_read\n");
+
+	amd_iommu_pc_get_set_reg_val(_GET_DEVID(event),
+				_GET_BANK(event), _GET_CNTR(event),
+				IOMMU_PC_COUNTER_REG, &count, false);
+
+	/* IOMMU pc counter register is only 48 bits */
+	count &= 0xFFFFFFFFFFFFULL;
+
+	prev_raw_count =  local64_read(&hwc->prev_count);
+	if (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
+					count) != prev_raw_count)
+		return;
+
+	delta = count - prev_raw_count;
+	local64_add(delta, &event->count);
+
+}
+
+static void perf_iommu_stop(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	u64 config;
+
+	pr_debug("perf: amd_iommu:perf_iommu_stop\n");
+
+	if (hwc->state & PERF_HES_UPTODATE)
+		return;
+
+	perf_iommu_disable_event(event);
+	WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
+	hwc->state |= PERF_HES_STOPPED;
+
+	if (hwc->state & PERF_HES_UPTODATE)
+		return;
+
+	config = hwc->config;
+	perf_iommu_read(event);
+	hwc->state |= PERF_HES_UPTODATE;
+}
+
+static int perf_iommu_add(struct perf_event *event, int flags)
+{
+	int retval;
+	struct perf_amd_iommu *perf_iommu =
+			container_of(event->pmu, struct perf_amd_iommu, pmu);
+
+	pr_debug("perf: amd_iommu:perf_iommu_add\n");
+	event->hw.state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
+
+	/* request an iommu bank/counter */
+	retval = get_next_avail_iommu_bnk_cntr(perf_iommu);
+	if (retval != -ENOSPC)
+		event->hw.extra_reg.reg = (u16)retval;
+	else
+		return retval;
+
+	if (flags & PERF_EF_START)
+		perf_iommu_start(event, PERF_EF_RELOAD);
+
+	return 0;
+}
+
+static void perf_iommu_del(struct perf_event *event, int flags)
+{
+	struct perf_amd_iommu *perf_iommu =
+			container_of(event->pmu, struct perf_amd_iommu, pmu);
+
+	pr_debug("perf: amd_iommu:perf_iommu_del\n");
+	perf_iommu_stop(event, PERF_EF_UPDATE);
+
+	/* clear the assigned iommu bank/counter */
+	clear_avail_iommu_bnk_cntr(perf_iommu,
+				     _GET_BANK(event),
+				     _GET_CNTR(event));
+
+	perf_event_update_userpage(event);
+}
+
+static __init int _init_events_attrs(struct perf_amd_iommu *perf_iommu)
+{
+	struct attribute **attrs;
+	struct attribute_group *attr_group;
+	int i = 0, j;
+
+	while (amd_iommu_v2_event_descs[i].attr.attr.name)
+		i++;
+
+	attr_group = kzalloc(sizeof(struct attribute *)
+		* (i + 1) + sizeof(*attr_group), GFP_KERNEL);
+	if (!attr_group)
+		return -ENOMEM;
+
+	attrs = (struct attribute **)(attr_group + 1);
+	for (j = 0; j < i; j++)
+		attrs[j] = &amd_iommu_v2_event_descs[j].attr.attr;
+
+	attr_group->name = "events";
+	attr_group->attrs = attrs;
+	perf_iommu->events_group = attr_group;
+
+	return 0;
+}
+
+static __init void amd_iommu_pc_exit(void)
+{
+	if (__perf_iommu.events_group != NULL) {
+		kfree(__perf_iommu.events_group);
+		__perf_iommu.events_group = NULL;
+	}
+}
+
+static __init int _init_perf_amd_iommu(
+	struct perf_amd_iommu *perf_iommu, char *name)
+{
+	int ret;
+
+	raw_spin_lock_init(&perf_iommu->lock);
+
+	/* Init format attributes */
+	perf_iommu->format_group = &amd_iommu_format_group;
+
+	/* Init cpumask attributes to only core 0 */
+	cpumask_set_cpu(0, &iommu_cpumask);
+	perf_iommu->cpumask_group = &amd_iommu_cpumask_group;
+
+	/* Init events attributes */
+	if (_init_events_attrs(perf_iommu) != 0)
+		pr_err("perf: amd_iommu: Only support raw events.\n");
+
+	/* Init null attributes */
+	perf_iommu->null_group = NULL;
+	perf_iommu->pmu.attr_groups = perf_iommu->attr_groups;
+
+	ret = perf_pmu_register(&perf_iommu->pmu, name, -1);
+	if (ret) {
+		pr_err("perf: amd_iommu: Failed to initialized.\n");
+		amd_iommu_pc_exit();
+	} else {
+		pr_info("perf: amd_iommu: Detected. (%d banks, %d counters/bank)\n",
+			amd_iommu_pc_get_max_banks(IOMMU_BASE_DEVID),
+			amd_iommu_pc_get_max_counters(IOMMU_BASE_DEVID));
+	}
+
+	return ret;
+}
+
+static struct perf_amd_iommu __perf_iommu = {
+	.pmu = {
+		.event_init	= perf_iommu_event_init,
+		.add		= perf_iommu_add,
+		.del		= perf_iommu_del,
+		.start		= perf_iommu_start,
+		.stop		= perf_iommu_stop,
+		.read		= perf_iommu_read,
+	},
+	.max_banks		= 0x00,
+	.max_counters		= 0x00,
+	.cntr_assign_mask	= 0ULL,
+	.format_group		= NULL,
+	.cpumask_group		= NULL,
+	.events_group		= NULL,
+	.null_group		= NULL,
+};
+
+static __init int amd_iommu_pc_init(void)
+{
+	/* Make sure the IOMMU PC resource is available */
+	if (!amd_iommu_pc_supported()) {
+		pr_err("perf: amd_iommu PMU not installed. No support!\n");
+		return -ENODEV;
+	}
+
+	_init_perf_amd_iommu(&__perf_iommu, "amd_iommu");
+
+	return 0;
+}
+
+device_initcall(amd_iommu_pc_init);
diff --git a/arch/x86/kernel/cpu/perf_event_amd_iommu.h b/arch/x86/kernel/cpu/perf_event_amd_iommu.h
new file mode 100644
index 0000000..845d173
--- /dev/null
+++ b/arch/x86/kernel/cpu/perf_event_amd_iommu.h
@@ -0,0 +1,40 @@
+/*
+ * Copyright (C) 2013 Advanced Micro Devices, Inc.
+ *
+ * Author: Steven Kinney <Steven.Kinney@amd.com>
+ * Author: Suravee Suthikulpanit <Suraveee.Suthikulpanit@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.
+ */
+
+#ifndef _PERF_EVENT_AMD_IOMMU_H_
+#define _PERF_EVENT_AMD_IOMMU_H_
+
+/* iommu pc mmio region register indexes */
+#define IOMMU_PC_COUNTER_REG			0x00
+#define IOMMU_PC_COUNTER_SRC_REG		0x08
+#define IOMMU_PC_PASID_MATCH_REG		0x10
+#define IOMMU_PC_DOMID_MATCH_REG		0x18
+#define IOMMU_PC_DEVID_MATCH_REG		0x20
+#define IOMMU_PC_COUNTER_REPORT_REG		0x28
+
+/* maximun specified bank/counters */
+#define PC_MAX_SPEC_BNKS			64
+#define PC_MAX_SPEC_CNTRS			16
+
+/* iommu pc reg masks*/
+#define IOMMU_BASE_DEVID			0x0000
+
+/* amd_iommu_init.c external support functions */
+extern bool amd_iommu_pc_supported(void);
+
+extern u8 amd_iommu_pc_get_max_banks(u16 devid);
+
+extern u8 amd_iommu_pc_get_max_counters(u16 devid);
+
+extern int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr,
+			u8 fxn, u64 *value, bool is_write);
+
+#endif /*_PERF_EVENT_AMD_IOMMU_H_*/
-- 
1.7.10.4

[PATCH 2/2 V4] perf/x86/amd: AMD IOMMU PC PERF uncore PMU implementation

[suravee.suthikulpanit-5C7GfCeVMHo]

From: Suravee Suthikulpanit <Suravee.Suthikulpanit-5C7GfCeVMHo@public.gmane.org>

Implement a perf PMU to handle IOMMU performance counters and events.
The PMU only supports counting mode (e.g. perf stat). Since the counters
are shared across all cores, the PMU is implemented as "system-wide" mode.

To invoke the AMD IOMMU PMU, issue a perf tool command such as:

./perf stat -a -e amd_iommu/<events>/ <command>
or
./perf stat -a -e amd_iommu/config=<config-data>,config1=<config1-data>/ <command>

For example:

./perf stat -a -e amd_iommu/mem_trans_total/ <command>

The resulting count will be how many IOMMU total peripheral memory
operations were performed during the command execution window.

Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit-5C7GfCeVMHo@public.gmane.org>
---
Changes from V3:
	- Add CONFIG_AMD_IOMMU in the Makefile
	- Add proper check for max_banks and max_counters

 arch/x86/kernel/cpu/Makefile               |    4 +
 arch/x86/kernel/cpu/perf_event_amd_iommu.c |  500 ++++++++++++++++++++++++++++
 arch/x86/kernel/cpu/perf_event_amd_iommu.h |   40 +++
 3 files changed, 544 insertions(+)
 create mode 100644 arch/x86/kernel/cpu/perf_event_amd_iommu.c
 create mode 100644 arch/x86/kernel/cpu/perf_event_amd_iommu.h

diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index b0684e4..47b56a7 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -31,11 +31,15 @@ obj-$(CONFIG_PERF_EVENTS)		+= perf_event.o
 
 ifdef CONFIG_PERF_EVENTS
 obj-$(CONFIG_CPU_SUP_AMD)		+= perf_event_amd.o perf_event_amd_uncore.o
+ifdef CONFIG_AMD_IOMMU
+obj-$(CONFIG_CPU_SUP_AMD)		+= perf_event_amd_iommu.o
+endif
 obj-$(CONFIG_CPU_SUP_INTEL)		+= perf_event_p6.o perf_event_knc.o perf_event_p4.o
 obj-$(CONFIG_CPU_SUP_INTEL)		+= perf_event_intel_lbr.o perf_event_intel_ds.o perf_event_intel.o
 obj-$(CONFIG_CPU_SUP_INTEL)		+= perf_event_intel_uncore.o
 endif
 
+
 obj-$(CONFIG_X86_MCE)			+= mcheck/
 obj-$(CONFIG_MTRR)			+= mtrr/
 
diff --git a/arch/x86/kernel/cpu/perf_event_amd_iommu.c b/arch/x86/kernel/cpu/perf_event_amd_iommu.c
new file mode 100644
index 0000000..a1199cd
--- /dev/null
+++ b/arch/x86/kernel/cpu/perf_event_amd_iommu.c
@@ -0,0 +1,500 @@
+/*
+ * Copyright (C) 2013 Advanced Micro Devices, Inc.
+ *
+ * Author: Steven Kinney <Steven.Kinney-5C7GfCeVMHo@public.gmane.org>
+ * Author: Suravee Suthikulpanit <Suraveee.Suthikulpanit-5C7GfCeVMHo@public.gmane.org>
+ *
+ * Perf: amd_iommu - AMD IOMMU Performance Counter PMU implementation
+ *
+ * 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.
+ */
+
+#include <linux/perf_event.h>
+#include <linux/module.h>
+#include <linux/cpumask.h>
+#include <linux/slab.h>
+
+#include "perf_event.h"
+#include "perf_event_amd_iommu.h"
+
+#define _GET_BANK(ev)       ((u8)(ev->hw.extra_reg.reg >> 8))
+#define _GET_CNTR(ev)       ((u8)(ev->hw.extra_reg.reg))
+
+/* iommu pmu config masks */
+#define _GET_CSOURCE(ev)    ((ev->hw.config & 0xFFULL))
+#define _GET_DEVID(ev)      ((ev->hw.config >> 8)  & 0xFFFFULL)
+#define _GET_PASID(ev)      ((ev->hw.config >> 24) & 0xFFFFULL)
+#define _GET_DOMID(ev)      ((ev->hw.config >> 40) & 0xFFFFULL)
+#define _GET_DEVID_MASK(ev) ((ev->hw.extra_reg.config)  & 0xFFFFULL)
+#define _GET_PASID_MASK(ev) ((ev->hw.extra_reg.config >> 16) & 0xFFFFULL)
+#define _GET_DOMID_MASK(ev) ((ev->hw.extra_reg.config >> 32) & 0xFFFFULL)
+
+static struct perf_amd_iommu __perf_iommu;
+
+struct perf_amd_iommu {
+	struct pmu pmu;
+	u8 max_banks;
+	u8 max_counters;
+	u64 cntr_assign_mask;
+	raw_spinlock_t lock;
+	const struct attribute_group *attr_groups[4];
+};
+
+#define format_group	attr_groups[0]
+#define cpumask_group	attr_groups[1]
+#define events_group	attr_groups[2]
+#define null_group	attr_groups[3]
+
+/*---------------------------------------------
+ * sysfs format attributes
+ *---------------------------------------------*/
+PMU_FORMAT_ATTR(csource,    "config:0-7");
+PMU_FORMAT_ATTR(devid,      "config:8-23");
+PMU_FORMAT_ATTR(pasid,      "config:24-39");
+PMU_FORMAT_ATTR(domid,      "config:40-55");
+PMU_FORMAT_ATTR(devid_mask, "config1:0-15");
+PMU_FORMAT_ATTR(pasid_mask, "config1:16-31");
+PMU_FORMAT_ATTR(domid_mask, "config1:32-47");
+
+static struct attribute *iommu_format_attrs[] = {
+	&format_attr_csource.attr,
+	&format_attr_devid.attr,
+	&format_attr_pasid.attr,
+	&format_attr_domid.attr,
+	&format_attr_devid_mask.attr,
+	&format_attr_pasid_mask.attr,
+	&format_attr_domid_mask.attr,
+	NULL,
+};
+
+static struct attribute_group amd_iommu_format_group = {
+	.name = "format",
+	.attrs = iommu_format_attrs,
+};
+
+/*---------------------------------------------
+ * sysfs events attributes
+ *---------------------------------------------*/
+struct amd_iommu_event_desc {
+	struct kobj_attribute attr;
+	const char *event;
+};
+
+static ssize_t _iommu_event_show(struct kobject *kobj,
+				struct kobj_attribute *attr, char *buf)
+{
+	struct amd_iommu_event_desc *event =
+		container_of(attr, struct amd_iommu_event_desc, attr);
+	return sprintf(buf, "%s\n", event->event);
+}
+
+#define AMD_IOMMU_EVENT_DESC(_name, _event)			\
+{								\
+	.attr  = __ATTR(_name, 0444, _iommu_event_show, NULL),	\
+	.event = _event,					\
+}
+
+static struct amd_iommu_event_desc amd_iommu_v2_event_descs[] = {
+	AMD_IOMMU_EVENT_DESC(mem_pass_untrans,        "csource=0x01"),
+	AMD_IOMMU_EVENT_DESC(mem_pass_pretrans,       "csource=0x02"),
+	AMD_IOMMU_EVENT_DESC(mem_pass_excl,           "csource=0x03"),
+	AMD_IOMMU_EVENT_DESC(mem_target_abort,        "csource=0x04"),
+	AMD_IOMMU_EVENT_DESC(mem_trans_total,         "csource=0x05"),
+	AMD_IOMMU_EVENT_DESC(mem_iommu_tlb_pte_hit,   "csource=0x06"),
+	AMD_IOMMU_EVENT_DESC(mem_iommu_tlb_pte_mis,   "csource=0x07"),
+	AMD_IOMMU_EVENT_DESC(mem_iommu_tlb_pde_hit,   "csource=0x08"),
+	AMD_IOMMU_EVENT_DESC(mem_iommu_tlb_pde_mis,   "csource=0x09"),
+	AMD_IOMMU_EVENT_DESC(mem_dte_hit,             "csource=0x0a"),
+	AMD_IOMMU_EVENT_DESC(mem_dte_mis,             "csource=0x0b"),
+	AMD_IOMMU_EVENT_DESC(page_tbl_read_tot,       "csource=0x0c"),
+	AMD_IOMMU_EVENT_DESC(page_tbl_read_nst,       "csource=0x0d"),
+	AMD_IOMMU_EVENT_DESC(page_tbl_read_gst,       "csource=0x0e"),
+	AMD_IOMMU_EVENT_DESC(int_dte_hit,             "csource=0x0f"),
+	AMD_IOMMU_EVENT_DESC(int_dte_mis,             "csource=0x10"),
+	AMD_IOMMU_EVENT_DESC(cmd_processed,           "csource=0x11"),
+	AMD_IOMMU_EVENT_DESC(cmd_processed_inv,       "csource=0x12"),
+	AMD_IOMMU_EVENT_DESC(tlb_inv,                 "csource=0x13"),
+	{ /* end: all zeroes */ },
+};
+
+/*---------------------------------------------
+ * sysfs cpumask attributes
+ *---------------------------------------------*/
+static cpumask_t iommu_cpumask;
+
+static ssize_t _iommu_cpumask_show(struct device *dev,
+				   struct device_attribute *attr,
+				   char *buf)
+{
+	int n = cpulist_scnprintf(buf, PAGE_SIZE - 2, &iommu_cpumask);
+	buf[n++] = '\n';
+	buf[n] = '\0';
+	return n;
+}
+static DEVICE_ATTR(cpumask, S_IRUGO, _iommu_cpumask_show, NULL);
+
+static struct attribute *iommu_cpumask_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL,
+};
+
+static struct attribute_group amd_iommu_cpumask_group = {
+	.attrs = iommu_cpumask_attrs,
+};
+
+/*---------------------------------------------*/
+
+static int get_next_avail_iommu_bnk_cntr(struct perf_amd_iommu *perf_iommu)
+{
+	unsigned long flags;
+	int shift, bank, cntr, retval;
+	int max_banks = perf_iommu->max_banks;
+	int max_cntrs = perf_iommu->max_counters;
+
+	raw_spin_lock_irqsave(&perf_iommu->lock, flags);
+
+	for (bank = 0, shift = 0; bank < max_banks; bank++) {
+		for (cntr = 0; cntr < max_cntrs; cntr++) {
+			shift = bank + (bank*3) + cntr;
+			if (perf_iommu->cntr_assign_mask & (1ULL<<shift)) {
+				continue;
+			} else {
+				perf_iommu->cntr_assign_mask |= (1ULL<<shift);
+				retval = ((u16)((u16)bank<<8) | (u8)(cntr));
+				goto out;
+			}
+		}
+	}
+	retval = -ENOSPC;
+out:
+	raw_spin_unlock_irqrestore(&perf_iommu->lock, flags);
+	return retval;
+}
+
+static int clear_avail_iommu_bnk_cntr(struct perf_amd_iommu *perf_iommu,
+					u8 bank, u8 cntr)
+{
+	unsigned long flags;
+	int max_banks, max_cntrs;
+	int shift = 0;
+
+	max_banks = perf_iommu->max_banks;
+	max_cntrs = perf_iommu->max_counters;
+
+	if ((bank > max_banks) || (cntr > max_cntrs))
+		return -EINVAL;
+
+	shift = bank + cntr + (bank*3);
+
+	raw_spin_lock_irqsave(&perf_iommu->lock, flags);
+	perf_iommu->cntr_assign_mask &= ~(1ULL<<shift);
+	raw_spin_unlock_irqrestore(&perf_iommu->lock, flags);
+
+	return 0;
+}
+
+static int perf_iommu_event_init(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct perf_amd_iommu *perf_iommu;
+	u64 config, config1;
+
+	/* test the event attr type check for PMU enumeration */
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	/*
+	 * IOMMU counters are shared across all cores.
+	 * Therefore, it does not support per-process mode.
+	 * Also, it does not support event sampling mode.
+	 */
+	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
+		return -EINVAL;
+
+	/* IOMMU counters do not have usr/os/guest/host bits */
+	if (event->attr.exclude_user || event->attr.exclude_kernel ||
+	    event->attr.exclude_host || event->attr.exclude_guest)
+		return -EINVAL;
+
+	if (event->cpu < 0)
+		return -EINVAL;
+
+	perf_iommu = &__perf_iommu;
+
+	if (event->pmu != &perf_iommu->pmu)
+		return -ENOENT;
+
+	if (perf_iommu) {
+		config = event->attr.config;
+		config1 = event->attr.config1;
+	} else {
+		return -EINVAL;
+	}
+
+	/* integrate with iommu base devid (0000), assume one iommu */
+	perf_iommu->max_banks =
+		amd_iommu_pc_get_max_banks(IOMMU_BASE_DEVID);
+	perf_iommu->max_counters =
+		amd_iommu_pc_get_max_counters(IOMMU_BASE_DEVID);
+	if ((perf_iommu->max_banks == 0) || (perf_iommu->max_counters == 0))
+		return -EINVAL;
+
+	/* update the hw_perf_event struct with the iommu config data */
+	hwc->config = config;
+	hwc->extra_reg.config = config1;
+
+	return 0;
+}
+
+static void perf_iommu_enable_event(struct perf_event *ev)
+{
+	u8 csource = _GET_CSOURCE(ev);
+	u16 devid = _GET_DEVID(ev);
+	u64 reg = 0ULL;
+
+	reg = csource;
+	amd_iommu_pc_get_set_reg_val(devid,
+			_GET_BANK(ev), _GET_CNTR(ev) ,
+			 IOMMU_PC_COUNTER_SRC_REG, &reg, true);
+
+	reg = 0ULL | devid | (_GET_DEVID_MASK(ev) << 32);
+	if (reg)
+		reg |= (1UL << 31);
+	amd_iommu_pc_get_set_reg_val(devid,
+			_GET_BANK(ev), _GET_CNTR(ev) ,
+			 IOMMU_PC_DEVID_MATCH_REG, &reg, true);
+
+	reg = 0ULL | _GET_PASID(ev) | (_GET_PASID_MASK(ev) << 32);
+	if (reg)
+		reg |= (1UL << 31);
+	amd_iommu_pc_get_set_reg_val(devid,
+			_GET_BANK(ev), _GET_CNTR(ev) ,
+			 IOMMU_PC_PASID_MATCH_REG, &reg, true);
+
+	reg = 0ULL | _GET_DOMID(ev) | (_GET_DOMID_MASK(ev) << 32);
+	if (reg)
+		reg |= (1UL << 31);
+	amd_iommu_pc_get_set_reg_val(devid,
+			_GET_BANK(ev), _GET_CNTR(ev) ,
+			 IOMMU_PC_DOMID_MATCH_REG, &reg, true);
+}
+
+static void perf_iommu_disable_event(struct perf_event *event)
+{
+	u64 reg = 0ULL;
+
+	amd_iommu_pc_get_set_reg_val(_GET_DEVID(event),
+			_GET_BANK(event), _GET_CNTR(event),
+			IOMMU_PC_COUNTER_SRC_REG, &reg, true);
+}
+
+static void perf_iommu_start(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+
+	pr_debug("perf: amd_iommu:perf_iommu_start\n");
+	if (WARN_ON_ONCE(!(hwc->state & PERF_HES_STOPPED)))
+		return;
+
+	WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));
+	hwc->state = 0;
+
+	if (flags & PERF_EF_RELOAD) {
+		u64 prev_raw_count =  local64_read(&hwc->prev_count);
+		amd_iommu_pc_get_set_reg_val(_GET_DEVID(event),
+				_GET_BANK(event), _GET_CNTR(event),
+				IOMMU_PC_COUNTER_REG, &prev_raw_count, true);
+	}
+
+	perf_iommu_enable_event(event);
+	perf_event_update_userpage(event);
+
+}
+
+static void perf_iommu_read(struct perf_event *event)
+{
+	u64 count = 0ULL;
+	u64 prev_raw_count = 0ULL;
+	u64 delta = 0ULL;
+	struct hw_perf_event *hwc = &event->hw;
+	pr_debug("perf: amd_iommu:perf_iommu_read\n");
+
+	amd_iommu_pc_get_set_reg_val(_GET_DEVID(event),
+				_GET_BANK(event), _GET_CNTR(event),
+				IOMMU_PC_COUNTER_REG, &count, false);
+
+	/* IOMMU pc counter register is only 48 bits */
+	count &= 0xFFFFFFFFFFFFULL;
+
+	prev_raw_count =  local64_read(&hwc->prev_count);
+	if (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
+					count) != prev_raw_count)
+		return;
+
+	delta = count - prev_raw_count;
+	local64_add(delta, &event->count);
+
+}
+
+static void perf_iommu_stop(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	u64 config;
+
+	pr_debug("perf: amd_iommu:perf_iommu_stop\n");
+
+	if (hwc->state & PERF_HES_UPTODATE)
+		return;
+
+	perf_iommu_disable_event(event);
+	WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
+	hwc->state |= PERF_HES_STOPPED;
+
+	if (hwc->state & PERF_HES_UPTODATE)
+		return;
+
+	config = hwc->config;
+	perf_iommu_read(event);
+	hwc->state |= PERF_HES_UPTODATE;
+}
+
+static int perf_iommu_add(struct perf_event *event, int flags)
+{
+	int retval;
+	struct perf_amd_iommu *perf_iommu =
+			container_of(event->pmu, struct perf_amd_iommu, pmu);
+
+	pr_debug("perf: amd_iommu:perf_iommu_add\n");
+	event->hw.state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
+
+	/* request an iommu bank/counter */
+	retval = get_next_avail_iommu_bnk_cntr(perf_iommu);
+	if (retval != -ENOSPC)
+		event->hw.extra_reg.reg = (u16)retval;
+	else
+		return retval;
+
+	if (flags & PERF_EF_START)
+		perf_iommu_start(event, PERF_EF_RELOAD);
+
+	return 0;
+}
+
+static void perf_iommu_del(struct perf_event *event, int flags)
+{
+	struct perf_amd_iommu *perf_iommu =
+			container_of(event->pmu, struct perf_amd_iommu, pmu);
+
+	pr_debug("perf: amd_iommu:perf_iommu_del\n");
+	perf_iommu_stop(event, PERF_EF_UPDATE);
+
+	/* clear the assigned iommu bank/counter */
+	clear_avail_iommu_bnk_cntr(perf_iommu,
+				     _GET_BANK(event),
+				     _GET_CNTR(event));
+
+	perf_event_update_userpage(event);
+}
+
+static __init int _init_events_attrs(struct perf_amd_iommu *perf_iommu)
+{
+	struct attribute **attrs;
+	struct attribute_group *attr_group;
+	int i = 0, j;
+
+	while (amd_iommu_v2_event_descs[i].attr.attr.name)
+		i++;
+
+	attr_group = kzalloc(sizeof(struct attribute *)
+		* (i + 1) + sizeof(*attr_group), GFP_KERNEL);
+	if (!attr_group)
+		return -ENOMEM;
+
+	attrs = (struct attribute **)(attr_group + 1);
+	for (j = 0; j < i; j++)
+		attrs[j] = &amd_iommu_v2_event_descs[j].attr.attr;
+
+	attr_group->name = "events";
+	attr_group->attrs = attrs;
+	perf_iommu->events_group = attr_group;
+
+	return 0;
+}
+
+static __init void amd_iommu_pc_exit(void)
+{
+	if (__perf_iommu.events_group != NULL) {
+		kfree(__perf_iommu.events_group);
+		__perf_iommu.events_group = NULL;
+	}
+}
+
+static __init int _init_perf_amd_iommu(
+	struct perf_amd_iommu *perf_iommu, char *name)
+{
+	int ret;
+
+	raw_spin_lock_init(&perf_iommu->lock);
+
+	/* Init format attributes */
+	perf_iommu->format_group = &amd_iommu_format_group;
+
+	/* Init cpumask attributes to only core 0 */
+	cpumask_set_cpu(0, &iommu_cpumask);
+	perf_iommu->cpumask_group = &amd_iommu_cpumask_group;
+
+	/* Init events attributes */
+	if (_init_events_attrs(perf_iommu) != 0)
+		pr_err("perf: amd_iommu: Only support raw events.\n");
+
+	/* Init null attributes */
+	perf_iommu->null_group = NULL;
+	perf_iommu->pmu.attr_groups = perf_iommu->attr_groups;
+
+	ret = perf_pmu_register(&perf_iommu->pmu, name, -1);
+	if (ret) {
+		pr_err("perf: amd_iommu: Failed to initialized.\n");
+		amd_iommu_pc_exit();
+	} else {
+		pr_info("perf: amd_iommu: Detected. (%d banks, %d counters/bank)\n",
+			amd_iommu_pc_get_max_banks(IOMMU_BASE_DEVID),
+			amd_iommu_pc_get_max_counters(IOMMU_BASE_DEVID));
+	}
+
+	return ret;
+}
+
+static struct perf_amd_iommu __perf_iommu = {
+	.pmu = {
+		.event_init	= perf_iommu_event_init,
+		.add		= perf_iommu_add,
+		.del		= perf_iommu_del,
+		.start		= perf_iommu_start,
+		.stop		= perf_iommu_stop,
+		.read		= perf_iommu_read,
+	},
+	.max_banks		= 0x00,
+	.max_counters		= 0x00,
+	.cntr_assign_mask	= 0ULL,
+	.format_group		= NULL,
+	.cpumask_group		= NULL,
+	.events_group		= NULL,
+	.null_group		= NULL,
+};
+
+static __init int amd_iommu_pc_init(void)
+{
+	/* Make sure the IOMMU PC resource is available */
+	if (!amd_iommu_pc_supported()) {
+		pr_err("perf: amd_iommu PMU not installed. No support!\n");
+		return -ENODEV;
+	}
+
+	_init_perf_amd_iommu(&__perf_iommu, "amd_iommu");
+
+	return 0;
+}
+
+device_initcall(amd_iommu_pc_init);
diff --git a/arch/x86/kernel/cpu/perf_event_amd_iommu.h b/arch/x86/kernel/cpu/perf_event_amd_iommu.h
new file mode 100644
index 0000000..845d173
--- /dev/null
+++ b/arch/x86/kernel/cpu/perf_event_amd_iommu.h
@@ -0,0 +1,40 @@
+/*
+ * Copyright (C) 2013 Advanced Micro Devices, Inc.
+ *
+ * Author: Steven Kinney <Steven.Kinney-5C7GfCeVMHo@public.gmane.org>
+ * Author: Suravee Suthikulpanit <Suraveee.Suthikulpanit-5C7GfCeVMHo@public.gmane.org>
+ *
+ * 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.
+ */
+
+#ifndef _PERF_EVENT_AMD_IOMMU_H_
+#define _PERF_EVENT_AMD_IOMMU_H_
+
+/* iommu pc mmio region register indexes */
+#define IOMMU_PC_COUNTER_REG			0x00
+#define IOMMU_PC_COUNTER_SRC_REG		0x08
+#define IOMMU_PC_PASID_MATCH_REG		0x10
+#define IOMMU_PC_DOMID_MATCH_REG		0x18
+#define IOMMU_PC_DEVID_MATCH_REG		0x20
+#define IOMMU_PC_COUNTER_REPORT_REG		0x28
+
+/* maximun specified bank/counters */
+#define PC_MAX_SPEC_BNKS			64
+#define PC_MAX_SPEC_CNTRS			16
+
+/* iommu pc reg masks*/
+#define IOMMU_BASE_DEVID			0x0000
+
+/* amd_iommu_init.c external support functions */
+extern bool amd_iommu_pc_supported(void);
+
+extern u8 amd_iommu_pc_get_max_banks(u16 devid);
+
+extern u8 amd_iommu_pc_get_max_counters(u16 devid);
+
+extern int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr,
+			u8 fxn, u64 *value, bool is_write);
+
+#endif /*_PERF_EVENT_AMD_IOMMU_H_*/
-- 
1.7.10.4

Re: [PATCH 0/2 V4] perf/x86/amd: IOMMU Performance Counter Support

[Suravee Suthikulanit]

Jorge,

Please let me know if you have any other questions/concerns.

Thanks,

Suravee

On 5/28/2013 5:45 PM, suravee.suthikulpanit@amd.com wrote:
> From: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
>
> These patches implement the AMD IOMMU Performance Counter functionality
> via custom perf PMU and implement static counting for various IOMMU
> translations.
>
>      1) Extend the AMD IOMMU initialization to include performance
>         counter enablement.
>
>      2) The perf AMD IOMMU PMU to manage performance counters, which
>         interface with the AMD IOMMU core driver.
>
> Steven L Kinney (1):
>    perf/x86/amd: Adding IOMMU PC resource management
>
> Suravee Suthikulpanit (1):
>    perf/x86/amd: AMD IOMMU PC PERF uncore PMU implementation
>
>   arch/x86/kernel/cpu/Makefile               |    4 +
>   arch/x86/kernel/cpu/perf_event_amd_iommu.c |  500 ++++++++++++++++++++++++++++
>   arch/x86/kernel/cpu/perf_event_amd_iommu.h |   40 +++
>   drivers/iommu/amd_iommu_init.c             |  140 +++++++-
>   drivers/iommu/amd_iommu_proto.h            |    7 +
>   drivers/iommu/amd_iommu_types.h            |   15 +-
>   6 files changed, 694 insertions(+), 12 deletions(-)
>   create mode 100644 arch/x86/kernel/cpu/perf_event_amd_iommu.c
>   create mode 100644 arch/x86/kernel/cpu/perf_event_amd_iommu.h
>

Re: [PATCH 0/2 V4] perf/x86/amd: IOMMU Performance Counter Support

[Suravee Suthikulanit]

Jorge,

Please let me know if you have any other questions/concerns.

Thanks,

Suravee

On 5/28/2013 5:45 PM, suravee.suthikulpanit-5C7GfCeVMHo@public.gmane.org wrote:
> From: Suravee Suthikulpanit <suravee.suthikulpanit-5C7GfCeVMHo@public.gmane.org>
>
> These patches implement the AMD IOMMU Performance Counter functionality
> via custom perf PMU and implement static counting for various IOMMU
> translations.
>
>      1) Extend the AMD IOMMU initialization to include performance
>         counter enablement.
>
>      2) The perf AMD IOMMU PMU to manage performance counters, which
>         interface with the AMD IOMMU core driver.
>
> Steven L Kinney (1):
>    perf/x86/amd: Adding IOMMU PC resource management
>
> Suravee Suthikulpanit (1):
>    perf/x86/amd: AMD IOMMU PC PERF uncore PMU implementation
>
>   arch/x86/kernel/cpu/Makefile               |    4 +
>   arch/x86/kernel/cpu/perf_event_amd_iommu.c |  500 ++++++++++++++++++++++++++++
>   arch/x86/kernel/cpu/perf_event_amd_iommu.h |   40 +++
>   drivers/iommu/amd_iommu_init.c             |  140 +++++++-
>   drivers/iommu/amd_iommu_proto.h            |    7 +
>   drivers/iommu/amd_iommu_types.h            |   15 +-
>   6 files changed, 694 insertions(+), 12 deletions(-)
>   create mode 100644 arch/x86/kernel/cpu/perf_event_amd_iommu.c
>   create mode 100644 arch/x86/kernel/cpu/perf_event_amd_iommu.h
>

Re: [PATCH 2/2 V4] perf/x86/amd: AMD IOMMU PC PERF uncore PMU implementation

[Peter Zijlstra]

On Tue, May 28, 2013 at 05:45:12PM -0500, suravee.suthikulpanit@amd.com wrote:
> +static void perf_iommu_start(struct perf_event *event, int flags)
> +{
> +	struct hw_perf_event *hwc = &event->hw;
> +
> +	pr_debug("perf: amd_iommu:perf_iommu_start\n");
> +	if (WARN_ON_ONCE(!(hwc->state & PERF_HES_STOPPED)))
> +		return;
> +
> +	WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));
> +	hwc->state = 0;
> +
> +	if (flags & PERF_EF_RELOAD) {
> +		u64 prev_raw_count =  local64_read(&hwc->prev_count);
> +		amd_iommu_pc_get_set_reg_val(_GET_DEVID(event),
> +				_GET_BANK(event), _GET_CNTR(event),
> +				IOMMU_PC_COUNTER_REG, &prev_raw_count, true);
> +	}
> +
> +	perf_iommu_enable_event(event);
> +	perf_event_update_userpage(event);
> +
> +}
> +
> +static void perf_iommu_read(struct perf_event *event)
> +{
> +	u64 count = 0ULL;
> +	u64 prev_raw_count = 0ULL;
> +	u64 delta = 0ULL;
> +	struct hw_perf_event *hwc = &event->hw;
> +	pr_debug("perf: amd_iommu:perf_iommu_read\n");
> +
> +	amd_iommu_pc_get_set_reg_val(_GET_DEVID(event),
> +				_GET_BANK(event), _GET_CNTR(event),
> +				IOMMU_PC_COUNTER_REG, &count, false);
> +
> +	/* IOMMU pc counter register is only 48 bits */
> +	count &= 0xFFFFFFFFFFFFULL;
> +
> +	prev_raw_count =  local64_read(&hwc->prev_count);
> +	if (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
> +					count) != prev_raw_count)
> +		return;
> +
> +	delta = count - prev_raw_count;
> +	local64_add(delta, &event->count);
> +
> +}

OK, so it looks like you have the event as a free-running counter; that
is, I don't see it being reset anywhere.

Combined with the fact that you're only 48bit wide, it looks like you
have a problem.

perf_iommu_read()'s delta is wrong. Imagine we've just wrapped and
prev_raw_count = 0xFFFFFFFFFFFFUL and count = 0x1. Then we do a 64bit
subtraction and end up with delta = 0xffff000000000002 or
18446462598732840962 instead of 2.

(also there's trailing whitespace in that function)

> +
> +static void perf_iommu_stop(struct perf_event *event, int flags)
> +{
> +	struct hw_perf_event *hwc = &event->hw;
> +	u64 config;
> +
> +	pr_debug("perf: amd_iommu:perf_iommu_stop\n");
> +
> +	if (hwc->state & PERF_HES_UPTODATE)
> +		return;
> +
> +	perf_iommu_disable_event(event);
> +	WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
> +	hwc->state |= PERF_HES_STOPPED;
> +
> +	if (hwc->state & PERF_HES_UPTODATE)
> +		return;
> +
> +	config = hwc->config;
> +	perf_iommu_read(event);
> +	hwc->state |= PERF_HES_UPTODATE;
> +}

Re: [PATCH 2/2 V4] perf/x86/amd: AMD IOMMU PC PERF uncore PMU implementation

[Peter Zijlstra]

On Tue, May 28, 2013 at 05:45:12PM -0500, suravee.suthikulpanit-5C7GfCeVMHo@public.gmane.org wrote:
> +static void perf_iommu_start(struct perf_event *event, int flags)
> +{
> +	struct hw_perf_event *hwc = &event->hw;
> +
> +	pr_debug("perf: amd_iommu:perf_iommu_start\n");
> +	if (WARN_ON_ONCE(!(hwc->state & PERF_HES_STOPPED)))
> +		return;
> +
> +	WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));
> +	hwc->state = 0;
> +
> +	if (flags & PERF_EF_RELOAD) {
> +		u64 prev_raw_count =  local64_read(&hwc->prev_count);
> +		amd_iommu_pc_get_set_reg_val(_GET_DEVID(event),
> +				_GET_BANK(event), _GET_CNTR(event),
> +				IOMMU_PC_COUNTER_REG, &prev_raw_count, true);
> +	}
> +
> +	perf_iommu_enable_event(event);
> +	perf_event_update_userpage(event);
> +
> +}
> +
> +static void perf_iommu_read(struct perf_event *event)
> +{
> +	u64 count = 0ULL;
> +	u64 prev_raw_count = 0ULL;
> +	u64 delta = 0ULL;
> +	struct hw_perf_event *hwc = &event->hw;
> +	pr_debug("perf: amd_iommu:perf_iommu_read\n");
> +
> +	amd_iommu_pc_get_set_reg_val(_GET_DEVID(event),
> +				_GET_BANK(event), _GET_CNTR(event),
> +				IOMMU_PC_COUNTER_REG, &count, false);
> +
> +	/* IOMMU pc counter register is only 48 bits */
> +	count &= 0xFFFFFFFFFFFFULL;
> +
> +	prev_raw_count =  local64_read(&hwc->prev_count);
> +	if (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
> +					count) != prev_raw_count)
> +		return;
> +
> +	delta = count - prev_raw_count;
> +	local64_add(delta, &event->count);
> +
> +}

OK, so it looks like you have the event as a free-running counter; that
is, I don't see it being reset anywhere.

Combined with the fact that you're only 48bit wide, it looks like you
have a problem.

perf_iommu_read()'s delta is wrong. Imagine we've just wrapped and
prev_raw_count = 0xFFFFFFFFFFFFUL and count = 0x1. Then we do a 64bit
subtraction and end up with delta = 0xffff000000000002 or
18446462598732840962 instead of 2.

(also there's trailing whitespace in that function)

> +
> +static void perf_iommu_stop(struct perf_event *event, int flags)
> +{
> +	struct hw_perf_event *hwc = &event->hw;
> +	u64 config;
> +
> +	pr_debug("perf: amd_iommu:perf_iommu_stop\n");
> +
> +	if (hwc->state & PERF_HES_UPTODATE)
> +		return;
> +
> +	perf_iommu_disable_event(event);
> +	WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
> +	hwc->state |= PERF_HES_STOPPED;
> +
> +	if (hwc->state & PERF_HES_UPTODATE)
> +		return;
> +
> +	config = hwc->config;
> +	perf_iommu_read(event);
> +	hwc->state |= PERF_HES_UPTODATE;
> +}