[PART2 PATCH v6 00/12] iommu/AMD: Introduce IOMMU AVIC support

[PART2 PATCH v6 00/12] iommu/AMD: Introduce IOMMU AVIC support

[Suravee Suthikulpanit]

CHANGES FROM V5
===============
Per Joerg:
    * Add documentation for amd_iommu_intr parameter.
    * Clean up early_amd_iommu_init() .
    * Do not update the tail pointer while processing ga_log.
      to avoid potential soft-lockup.

Per Radim:
    * Fix vm_id logic.
    * Refactor avic_set_running().
    * Use per-vcpu list of struct amd_ir_data instead of
      struct amd_iommu_pi_data. This result in changes in
      svm_update_pi_irte().

Misc:
    * Clean up amd_iommu_update_ga()
    * Other minor clean ups.

GITHUB
======
Latest git tree can be found at:
    http://github.com/ssuthiku/linux.git    avic_part2_v6

OVERVIEW
========
This patch set is the second part of the two-part patch series to introduce
the new AMD Advance Virtual Interrupt Controller (AVIC) support.

In addition to the SVM AVIC, AMD IOMMU also extends the AVIC capability
to allow I/O interrupts injection directly into the virtualized guest
local APIC without the need for hypervisor intervention.

This patch series introduces a new hardware interrupt remapping (IR) mode
in AMD IOMMU driver, the Guest Virtual APIC (GA) mode. This is in contrast
to the existing "legacy" mode. The IR mode can be specified with a new
kernel parameter:

    amd_iommu_guest_ir=[vapic (default) | legacy]

When enabling GA mode, the AMD IOMMU driver will configure device interrupt
remapping in GA mode when possible (i.e. SVM AVIC must be enabled, and if
the interrupt types are supported). Otherewise, the driver will fallback
to using the legacy IR mode.

This patch series also introduces new interfaces between SVM and IOMMU
to allow:
  * SVM driver to communicate to IOMMU with updated vcpu scheduling
    information.
  * IOMMU driver to notify SVM driver to schedule vcpu on to physical core
    handle IOMMU GALog entry.

DOCUMENTATIONS
==============
More information about SVM AVIC can be found in the
AMD64 Architecture Programmer’s Manual Volume 2 - System Programming.

    http://support.amd.com/TechDocs/24593.pdf

More information about IOMMU AVIC can be found int the
AMD I/O Virtualization Technology (IOMMU) Specification - Rev 2.62.

    http://support.amd.com/TechDocs/48882_IOMMU.pdf

Any feedback and comments are very much appreciated.

Thank you,
Suravee

Suravee Suthikulpanit (12):
  iommu/amd: Detect and enable guest vAPIC support
  iommu/amd: Move and introduce new IRTE-related unions and structures
  iommu/amd: Introduce interrupt remapping ops structure
  iommu/amd: Add support for multiple IRTE formats
  iommu/amd: Detect and initialize guest vAPIC log
  iommu/amd: Adding GALOG interrupt handler
  iommu/amd: Introduce amd_iommu_update_ga()
  iommu/amd: Implements irq_set_vcpu_affinity() hook to setup vapic mode
    for pass-through devices
  iommu/amd: Enable vAPIC interrupt remapping mode by default
  svm: Introduces AVIC per-VM ID
  svm: Introduce AMD IOMMU avic_ga_log_notifier
  svm: Implements update_pi_irte hook to setup posted interrupt

 Documentation/kernel-parameters.txt |   9 +
 arch/x86/include/asm/kvm_host.h     |   2 +
 arch/x86/kvm/svm.c                  | 371 +++++++++++++++++++++++++--
 drivers/iommu/amd_iommu.c           | 487 +++++++++++++++++++++++++++++++-----
 drivers/iommu/amd_iommu_init.c      | 181 +++++++++++++-
 drivers/iommu/amd_iommu_proto.h     |   1 +
 drivers/iommu/amd_iommu_types.h     | 136 ++++++++++
 include/linux/amd-iommu.h           |  56 ++++-
 8 files changed, 1154 insertions(+), 89 deletions(-)

-- 
1.9.1

[PART2 PATCH v6 01/12] iommu/amd: Detect and enable guest vAPIC support

[Suravee Suthikulpanit]

This patch introduces a new IOMMU driver parameter, amd_iommu_guest_ir,
which can be used to specify different interrupt remapping mode for
passthrough devices to VM guest:
    * legacy: Legacy interrupt remapping (w/ 32-bit IRTE)
    * vapic : Guest vAPIC interrupt remapping (w/ GA mode 128-bit IRTE)

Note that in vapic mode, it can also supports legacy interrupt remapping
for non-passthrough devices with the 128-bit IRTE.

Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
---
 Documentation/kernel-parameters.txt |  9 +++++
 drivers/iommu/amd_iommu_init.c      | 71 +++++++++++++++++++++++++++++++++----
 drivers/iommu/amd_iommu_proto.h     |  1 +
 drivers/iommu/amd_iommu_types.h     | 24 +++++++++++++
 4 files changed, 99 insertions(+), 6 deletions(-)

diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
index 17e33db..66c8f4b 100644
--- a/Documentation/kernel-parameters.txt
+++ b/Documentation/kernel-parameters.txt
@@ -460,6 +460,15 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
 			driver will print ACPI tables for AMD IOMMU during
 			IOMMU initialization.
 
+	amd_iommu_intr=	[HW,X86-64]
+			Specifies one of the following AMD IOMMU interrupt
+			remapping modes:
+			legacy     - Use legacy interrupt remapping mode.
+			vapic      - Use virtual APIC mode, which allows IOMMU
+			             to inject interrupts directly into guest.
+			             This mode requires kvm-amd.avic=1.
+			             (Default when IOMMU HW support is present.)
+
 	amijoy.map=	[HW,JOY] Amiga joystick support
 			Map of devices attached to JOY0DAT and JOY1DAT
 			Format: <a>,<b>
diff --git a/drivers/iommu/amd_iommu_init.c b/drivers/iommu/amd_iommu_init.c
index 59741ea..c3afd86 100644
--- a/drivers/iommu/amd_iommu_init.c
+++ b/drivers/iommu/amd_iommu_init.c
@@ -145,6 +145,8 @@ struct ivmd_header {
 bool amd_iommu_dump;
 bool amd_iommu_irq_remap __read_mostly;
 
+int amd_iommu_guest_ir;
+
 static bool amd_iommu_detected;
 static bool __initdata amd_iommu_disabled;
 static int amd_iommu_target_ivhd_type;
@@ -1258,6 +1260,8 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
 		else
 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
+		if (((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
+			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
 		break;
 	case 0x11:
 	case 0x40:
@@ -1265,6 +1269,8 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
 		else
 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
+		if (((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0))
+			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
 		break;
 	default:
 		return -EINVAL;
@@ -1488,6 +1494,14 @@ static int iommu_init_pci(struct amd_iommu *iommu)
 	if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
 		return -ENOMEM;
 
+	/* Note: We have already checked GASup from IVRS table.
+	 *       Now, we need to make sure that GAMSup is set.
+	 */
+	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
+	    !iommu_feature(iommu, FEATURE_GAM_VAPIC))
+		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
+
+
 	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
 		amd_iommu_np_cache = true;
 
@@ -1545,16 +1559,24 @@ static void print_iommu_info(void)
 			dev_name(&iommu->dev->dev), iommu->cap_ptr);
 
 		if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
-			pr_info("AMD-Vi:  Extended features: ");
+			pr_info("AMD-Vi: Extended features (%#llx):\n",
+				iommu->features);
 			for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
 				if (iommu_feature(iommu, (1ULL << i)))
 					pr_cont(" %s", feat_str[i]);
 			}
+
+			if (iommu->features & FEATURE_GAM_VAPIC)
+				pr_cont(" GA_vAPIC");
+
 			pr_cont("\n");
 		}
 	}
-	if (irq_remapping_enabled)
+	if (irq_remapping_enabled) {
 		pr_info("AMD-Vi: Interrupt remapping enabled\n");
+		if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
+			pr_info("AMD-Vi: virtual APIC enabled\n");
+	}
 }
 
 static int __init amd_iommu_init_pci(void)
@@ -1862,6 +1884,22 @@ static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
 			       iommu->stored_addr_lo | 1);
 }
 
+static void iommu_enable_ga(struct amd_iommu *iommu)
+{
+#ifdef CONFIG_IRQ_REMAP
+	switch (amd_iommu_guest_ir) {
+	case AMD_IOMMU_GUEST_IR_VAPIC:
+		iommu_feature_enable(iommu, CONTROL_GAM_EN);
+		/* Fall through */
+	case AMD_IOMMU_GUEST_IR_LEGACY_GA:
+		iommu_feature_enable(iommu, CONTROL_GA_EN);
+		break;
+	default:
+		break;
+	}
+#endif
+}
+
 /*
  * This function finally enables all IOMMUs found in the system after
  * they have been initialized
@@ -1877,6 +1915,7 @@ static void early_enable_iommus(void)
 		iommu_enable_command_buffer(iommu);
 		iommu_enable_event_buffer(iommu);
 		iommu_set_exclusion_range(iommu);
+		iommu_enable_ga(iommu);
 		iommu_enable(iommu);
 		iommu_flush_all_caches(iommu);
 	}
@@ -2059,7 +2098,7 @@ static int __init early_amd_iommu_init(void)
 	struct acpi_table_header *ivrs_base;
 	acpi_size ivrs_size;
 	acpi_status status;
-	int i, ret = 0;
+	int i, remap_cache_sz, ret = 0;
 
 	if (!amd_iommu_detected)
 		return -ENODEV;
@@ -2157,10 +2196,14 @@ static int __init early_amd_iommu_init(void)
 		 * remapping tables.
 		 */
 		ret = -ENOMEM;
+		if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
+			remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
+		else
+			remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
 		amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
-				MAX_IRQS_PER_TABLE * sizeof(u32),
-				IRQ_TABLE_ALIGNMENT,
-				0, NULL);
+							remap_cache_sz,
+							IRQ_TABLE_ALIGNMENT,
+							0, NULL);
 		if (!amd_iommu_irq_cache)
 			goto out;
 
@@ -2413,6 +2456,21 @@ static int __init parse_amd_iommu_dump(char *str)
 	return 1;
 }
 
+static int __init parse_amd_iommu_intr(char *str)
+{
+	for (; *str; ++str) {
+		if (strncmp(str, "legacy", 6) == 0) {
+			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
+			break;
+		}
+		if (strncmp(str, "vapic", 5) == 0) {
+			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
+			break;
+		}
+	}
+	return 1;
+}
+
 static int __init parse_amd_iommu_options(char *str)
 {
 	for (; *str; ++str) {
@@ -2521,6 +2579,7 @@ static int __init parse_ivrs_acpihid(char *str)
 
 __setup("amd_iommu_dump",	parse_amd_iommu_dump);
 __setup("amd_iommu=",		parse_amd_iommu_options);
+__setup("amd_iommu_intr=",	parse_amd_iommu_intr);
 __setup("ivrs_ioapic",		parse_ivrs_ioapic);
 __setup("ivrs_hpet",		parse_ivrs_hpet);
 __setup("ivrs_acpihid",		parse_ivrs_acpihid);
diff --git a/drivers/iommu/amd_iommu_proto.h b/drivers/iommu/amd_iommu_proto.h
index 0bd9eb3..faa3b48 100644
--- a/drivers/iommu/amd_iommu_proto.h
+++ b/drivers/iommu/amd_iommu_proto.h
@@ -38,6 +38,7 @@ extern int amd_iommu_enable(void);
 extern void amd_iommu_disable(void);
 extern int amd_iommu_reenable(int);
 extern int amd_iommu_enable_faulting(void);
+extern int amd_iommu_guest_ir;
 
 /* IOMMUv2 specific functions */
 struct iommu_domain;
diff --git a/drivers/iommu/amd_iommu_types.h b/drivers/iommu/amd_iommu_types.h
index 590956a..25f939b 100644
--- a/drivers/iommu/amd_iommu_types.h
+++ b/drivers/iommu/amd_iommu_types.h
@@ -92,6 +92,7 @@
 #define FEATURE_GA		(1ULL<<7)
 #define FEATURE_HE		(1ULL<<8)
 #define FEATURE_PC		(1ULL<<9)
+#define FEATURE_GAM_VAPIC	(1ULL<<21)
 
 #define FEATURE_PASID_SHIFT	32
 #define FEATURE_PASID_MASK	(0x1fULL << FEATURE_PASID_SHIFT)
@@ -146,6 +147,8 @@
 #define CONTROL_PPFINT_EN       0x0eULL
 #define CONTROL_PPR_EN          0x0fULL
 #define CONTROL_GT_EN           0x10ULL
+#define CONTROL_GA_EN           0x11ULL
+#define CONTROL_GAM_EN          0x19ULL
 
 #define CTRL_INV_TO_MASK	(7 << CONTROL_INV_TIMEOUT)
 #define CTRL_INV_TO_NONE	0
@@ -329,6 +332,12 @@
 #define IOMMU_CAP_NPCACHE 26
 #define IOMMU_CAP_EFR     27
 
+/* IOMMU Feature Reporting Field (for IVHD type 10h */
+#define IOMMU_FEAT_GASUP_SHIFT	6
+
+/* IOMMU Extended Feature Register (EFR) */
+#define IOMMU_EFR_GASUP_SHIFT	7
+
 #define MAX_DOMAIN_ID 65536
 
 /* Protection domain flags */
@@ -682,4 +691,19 @@ static inline int get_hpet_devid(int id)
 	return -EINVAL;
 }
 
+enum amd_iommu_intr_mode_type {
+	AMD_IOMMU_GUEST_IR_LEGACY,
+
+	/* This mode is not visible to users. It is used when
+	 * we cannot fully enable vAPIC and fallback to only support
+	 * legacy interrupt remapping via 128-bit IRTE.
+	 */
+	AMD_IOMMU_GUEST_IR_LEGACY_GA,
+	AMD_IOMMU_GUEST_IR_VAPIC,
+};
+
+#define AMD_IOMMU_GUEST_IR_GA(x)	(x == AMD_IOMMU_GUEST_IR_VAPIC || \
+					 x == AMD_IOMMU_GUEST_IR_LEGACY_GA)
+
+#define AMD_IOMMU_GUEST_IR_VAPIC(x)	(x == AMD_IOMMU_GUEST_IR_VAPIC)
 #endif /* _ASM_X86_AMD_IOMMU_TYPES_H */
-- 
1.9.1

[PART2 PATCH v6 02/12] iommu/amd: Move and introduce new IRTE-related unions and structures

[Suravee Suthikulpanit]

Move existing unions and structs for accessing/managing IRTE to a proper
header file. This is mainly to simplify variable declarations in subsequent
patches.

Besides, this patch also introduces new struct irte_ga for the new
128-bit IRTE format.

Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
---
 drivers/iommu/amd_iommu.c       | 28 ---------------
 drivers/iommu/amd_iommu_types.h | 76 +++++++++++++++++++++++++++++++++++++++++
 2 files changed, 76 insertions(+), 28 deletions(-)

diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
index 634f636..ac2962f 100644
--- a/drivers/iommu/amd_iommu.c
+++ b/drivers/iommu/amd_iommu.c
@@ -3693,34 +3693,6 @@ EXPORT_SYMBOL(amd_iommu_device_info);
  *
  *****************************************************************************/
 
-union irte {
-	u32 val;
-	struct {
-		u32 valid	: 1,
-		    no_fault	: 1,
-		    int_type	: 3,
-		    rq_eoi	: 1,
-		    dm		: 1,
-		    rsvd_1	: 1,
-		    destination	: 8,
-		    vector	: 8,
-		    rsvd_2	: 8;
-	} fields;
-};
-
-struct irq_2_irte {
-	u16 devid; /* Device ID for IRTE table */
-	u16 index; /* Index into IRTE table*/
-};
-
-struct amd_ir_data {
-	struct irq_2_irte			irq_2_irte;
-	union irte				irte_entry;
-	union {
-		struct msi_msg			msi_entry;
-	};
-};
-
 static struct irq_chip amd_ir_chip;
 
 #define DTE_IRQ_PHYS_ADDR_MASK	(((1ULL << 45)-1) << 6)
diff --git a/drivers/iommu/amd_iommu_types.h b/drivers/iommu/amd_iommu_types.h
index 25f939b..c37c5c4 100644
--- a/drivers/iommu/amd_iommu_types.h
+++ b/drivers/iommu/amd_iommu_types.h
@@ -22,6 +22,7 @@
 
 #include <linux/types.h>
 #include <linux/mutex.h>
+#include <linux/msi.h>
 #include <linux/list.h>
 #include <linux/spinlock.h>
 #include <linux/pci.h>
@@ -706,4 +707,79 @@ enum amd_iommu_intr_mode_type {
 					 x == AMD_IOMMU_GUEST_IR_LEGACY_GA)
 
 #define AMD_IOMMU_GUEST_IR_VAPIC(x)	(x == AMD_IOMMU_GUEST_IR_VAPIC)
+
+union irte {
+	u32 val;
+	struct {
+		u32 valid	: 1,
+		    no_fault	: 1,
+		    int_type	: 3,
+		    rq_eoi	: 1,
+		    dm		: 1,
+		    rsvd_1	: 1,
+		    destination	: 8,
+		    vector	: 8,
+		    rsvd_2	: 8;
+	} fields;
+};
+
+union irte_ga_lo {
+	u64 val;
+
+	/* For int remapping */
+	struct {
+		u64 valid	: 1,
+		    no_fault	: 1,
+		    /* ------ */
+		    int_type	: 3,
+		    rq_eoi	: 1,
+		    dm		: 1,
+		    /* ------ */
+		    guest_mode	: 1,
+		    destination	: 8,
+		    rsvd	: 48;
+	} fields_remap;
+
+	/* For guest vAPIC */
+	struct {
+		u64 valid	: 1,
+		    no_fault	: 1,
+		    /* ------ */
+		    ga_log_intr	: 1,
+		    rsvd1	: 3,
+		    is_run	: 1,
+		    /* ------ */
+		    guest_mode	: 1,
+		    destination	: 8,
+		    rsvd2	: 16,
+		    ga_tag	: 32;
+	} fields_vapic;
+};
+
+union irte_ga_hi {
+	u64 val;
+	struct {
+		u64 vector	: 8,
+		    rsvd_1	: 4,
+		    ga_root_ptr	: 40,
+		    rsvd_2	: 12;
+	} fields;
+};
+
+struct irte_ga {
+	union irte_ga_lo lo;
+	union irte_ga_hi hi;
+};
+
+struct irq_2_irte {
+	u16 devid; /* Device ID for IRTE table */
+	u16 index; /* Index into IRTE table*/
+};
+
+struct amd_ir_data {
+	struct irq_2_irte irq_2_irte;
+	union irte irte_entry;
+	struct msi_msg msi_entry;
+};
+
 #endif /* _ASM_X86_AMD_IOMMU_TYPES_H */
-- 
1.9.1

[PART2 PATCH v6 03/12] iommu/amd: Introduce interrupt remapping ops structure

[Suravee Suthikulpanit]

Currently, IOMMU support two interrupt remapping table entry formats,
32-bit (legacy) and 128-bit (GA). The spec also implies that it might
support additional modes/formats in the future.

So, this patch introduces the new struct amd_irte_ops, which allows
the same code to work with different irte formats by providing hooks
for various operations on an interrupt remapping table entry.

Suggested-by: Joerg Roedel <joro@8bytes.org>
Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
---
 drivers/iommu/amd_iommu.c       | 193 ++++++++++++++++++++++++++++++++++++++--
 drivers/iommu/amd_iommu_types.h |  20 +++++
 2 files changed, 208 insertions(+), 5 deletions(-)

diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
index ac2962f..139ea8b 100644
--- a/drivers/iommu/amd_iommu.c
+++ b/drivers/iommu/amd_iommu.c
@@ -3826,11 +3826,13 @@ out:
 	return index;
 }
 
-static int modify_irte(u16 devid, int index, union irte irte)
+static int modify_irte_ga(u16 devid, int index, struct irte_ga *irte)
 {
 	struct irq_remap_table *table;
 	struct amd_iommu *iommu;
 	unsigned long flags;
+	struct irte_ga *entry;
+	struct irte_ga tmp;
 
 	iommu = amd_iommu_rlookup_table[devid];
 	if (iommu == NULL)
@@ -3841,7 +3843,40 @@ static int modify_irte(u16 devid, int index, union irte irte)
 		return -ENOMEM;
 
 	spin_lock_irqsave(&table->lock, flags);
-	table->table[index] = irte.val;
+
+	entry = (struct irte_ga *)table->table;
+	entry = &entry[index];
+	memcpy(&tmp, entry, sizeof(struct irte_ga));
+	entry->lo.fields_remap.valid = 0;
+	entry->hi.val = irte->hi.val;
+	entry->hi.fields.ga_root_ptr = tmp.hi.fields.ga_root_ptr;
+	entry->lo.val = irte->lo.val;
+	entry->lo.fields_remap.valid = 1;
+
+	spin_unlock_irqrestore(&table->lock, flags);
+
+	iommu_flush_irt(iommu, devid);
+	iommu_completion_wait(iommu);
+
+	return 0;
+}
+
+static int modify_irte(u16 devid, int index, union irte *irte)
+{
+	struct irq_remap_table *table;
+	struct amd_iommu *iommu;
+	unsigned long flags;
+
+	iommu = amd_iommu_rlookup_table[devid];
+	if (iommu == NULL)
+		return -EINVAL;
+
+	table = get_irq_table(devid, false);
+	if (!table)
+		return -ENOMEM;
+
+	spin_lock_irqsave(&table->lock, flags);
+	table->table[index] = irte->val;
 	spin_unlock_irqrestore(&table->lock, flags);
 
 	iommu_flush_irt(iommu, devid);
@@ -3872,6 +3907,134 @@ static void free_irte(u16 devid, int index)
 	iommu_completion_wait(iommu);
 }
 
+static void irte_prepare(void *entry,
+			 u32 delivery_mode, u32 dest_mode,
+			 u8 vector, u32 dest_apicid)
+{
+	union irte *irte = (union irte *) entry;
+
+	irte->val                = 0;
+	irte->fields.vector      = vector;
+	irte->fields.int_type    = delivery_mode;
+	irte->fields.destination = dest_apicid;
+	irte->fields.dm          = dest_mode;
+	irte->fields.valid       = 1;
+}
+
+static void irte_ga_prepare(void *entry,
+			    u32 delivery_mode, u32 dest_mode,
+			    u8 vector, u32 dest_apicid)
+{
+	struct irte_ga *irte = (struct irte_ga *) entry;
+
+	irte->lo.val                      = 0;
+	irte->hi.val                      = 0;
+	irte->lo.fields_remap.guest_mode  = 0;
+	irte->lo.fields_remap.int_type    = delivery_mode;
+	irte->lo.fields_remap.dm          = dest_mode;
+	irte->hi.fields.vector            = vector;
+	irte->lo.fields_remap.destination = dest_apicid;
+	irte->lo.fields_remap.valid       = 1;
+}
+
+static void irte_activate(void *entry, u16 devid, u16 index)
+{
+	union irte *irte = (union irte *) entry;
+
+	irte->fields.valid = 1;
+	modify_irte(devid, index, irte);
+}
+
+static void irte_ga_activate(void *entry, u16 devid, u16 index)
+{
+	struct irte_ga *irte = (struct irte_ga *) entry;
+
+	irte->lo.fields_remap.valid = 1;
+	modify_irte_ga(devid, index, irte);
+}
+
+static void irte_deactivate(void *entry, u16 devid, u16 index)
+{
+	union irte *irte = (union irte *) entry;
+
+	irte->fields.valid = 0;
+	modify_irte(devid, index, irte);
+}
+
+static void irte_ga_deactivate(void *entry, u16 devid, u16 index)
+{
+	struct irte_ga *irte = (struct irte_ga *) entry;
+
+	irte->lo.fields_remap.valid = 0;
+	modify_irte_ga(devid, index, irte);
+}
+
+static void irte_set_affinity(void *entry, u16 devid, u16 index,
+			      u8 vector, u32 dest_apicid)
+{
+	union irte *irte = (union irte *) entry;
+
+	irte->fields.vector = vector;
+	irte->fields.destination = dest_apicid;
+	modify_irte(devid, index, irte);
+}
+
+static void irte_ga_set_affinity(void *entry, u16 devid, u16 index,
+				 u8 vector, u32 dest_apicid)
+{
+	struct irte_ga *irte = (struct irte_ga *) entry;
+
+	irte->hi.fields.vector = vector;
+	irte->lo.fields_remap.destination = dest_apicid;
+	irte->lo.fields_remap.guest_mode = 0;
+	modify_irte_ga(devid, index, irte);
+}
+
+static void irte_set_allocated(struct irq_remap_table *table, int index)
+{
+	table->table[index] = IRTE_ALLOCATED;
+}
+
+static void irte_ga_set_allocated(struct irq_remap_table *table, int index)
+{
+	struct irte_ga *ptr = (struct irte_ga *)table->table;
+	struct irte_ga *irte = &ptr[index];
+
+	memset(&irte->lo.val, 0, sizeof(u64));
+	memset(&irte->hi.val, 0, sizeof(u64));
+	irte->hi.fields.vector = 0xff;
+}
+
+static bool irte_is_allocated(struct irq_remap_table *table, int index)
+{
+	union irte *ptr = (union irte *)table->table;
+	union irte *irte = &ptr[index];
+
+	return irte->val != 0;
+}
+
+static bool irte_ga_is_allocated(struct irq_remap_table *table, int index)
+{
+	struct irte_ga *ptr = (struct irte_ga *)table->table;
+	struct irte_ga *irte = &ptr[index];
+
+	return irte->hi.fields.vector != 0;
+}
+
+static void irte_clear_allocated(struct irq_remap_table *table, int index)
+{
+	table->table[index] = 0;
+}
+
+static void irte_ga_clear_allocated(struct irq_remap_table *table, int index)
+{
+	struct irte_ga *ptr = (struct irte_ga *)table->table;
+	struct irte_ga *irte = &ptr[index];
+
+	memset(&irte->lo.val, 0, sizeof(u64));
+	memset(&irte->hi.val, 0, sizeof(u64));
+}
+
 static int get_devid(struct irq_alloc_info *info)
 {
 	int devid = -1;
@@ -3999,6 +4162,26 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data,
 	}
 }
 
+struct amd_irte_ops irte_32_ops = {
+	.prepare = irte_prepare,
+	.activate = irte_activate,
+	.deactivate = irte_deactivate,
+	.set_affinity = irte_set_affinity,
+	.set_allocated = irte_set_allocated,
+	.is_allocated = irte_is_allocated,
+	.clear_allocated = irte_clear_allocated,
+};
+
+struct amd_irte_ops irte_128_ops = {
+	.prepare = irte_ga_prepare,
+	.activate = irte_ga_activate,
+	.deactivate = irte_ga_deactivate,
+	.set_affinity = irte_ga_set_affinity,
+	.set_allocated = irte_ga_set_allocated,
+	.is_allocated = irte_ga_is_allocated,
+	.clear_allocated = irte_ga_clear_allocated,
+};
+
 static int irq_remapping_alloc(struct irq_domain *domain, unsigned int virq,
 			       unsigned int nr_irqs, void *arg)
 {
@@ -4104,7 +4287,7 @@ static void irq_remapping_activate(struct irq_domain *domain,
 	struct amd_ir_data *data = irq_data->chip_data;
 	struct irq_2_irte *irte_info = &data->irq_2_irte;
 
-	modify_irte(irte_info->devid, irte_info->index, data->irte_entry);
+	modify_irte(irte_info->devid, irte_info->index, &data->irte_entry);
 }
 
 static void irq_remapping_deactivate(struct irq_domain *domain,
@@ -4115,7 +4298,7 @@ static void irq_remapping_deactivate(struct irq_domain *domain,
 	union irte entry;
 
 	entry.val = 0;
-	modify_irte(irte_info->devid, irte_info->index, data->irte_entry);
+	modify_irte(irte_info->devid, irte_info->index, &data->irte_entry);
 }
 
 static struct irq_domain_ops amd_ir_domain_ops = {
@@ -4144,7 +4327,7 @@ static int amd_ir_set_affinity(struct irq_data *data,
 	 */
 	ir_data->irte_entry.fields.vector = cfg->vector;
 	ir_data->irte_entry.fields.destination = cfg->dest_apicid;
-	modify_irte(irte_info->devid, irte_info->index, ir_data->irte_entry);
+	modify_irte(irte_info->devid, irte_info->index, &ir_data->irte_entry);
 
 	/*
 	 * After this point, all the interrupts will start arriving
diff --git a/drivers/iommu/amd_iommu_types.h b/drivers/iommu/amd_iommu_types.h
index c37c5c4..d57c56d 100644
--- a/drivers/iommu/amd_iommu_types.h
+++ b/drivers/iommu/amd_iommu_types.h
@@ -410,6 +410,7 @@ struct amd_iommu_fault {
 
 struct iommu_domain;
 struct irq_domain;
+struct amd_irte_ops;
 
 /*
  * This structure contains generic data for  IOMMU protection domains
@@ -534,6 +535,8 @@ struct amd_iommu {
 #ifdef CONFIG_IRQ_REMAP
 	struct irq_domain *ir_domain;
 	struct irq_domain *msi_domain;
+
+	struct amd_irte_ops *irte_ops;
 #endif
 };
 
@@ -780,6 +783,23 @@ struct amd_ir_data {
 	struct irq_2_irte irq_2_irte;
 	union irte irte_entry;
 	struct msi_msg msi_entry;
+	void *entry;    /* Pointer to union irte or struct irte_ga */
+};
+
+struct amd_irte_ops {
+	void (*prepare)(void *, u32, u32, u8, u32);
+	void (*activate)(void *, u16, u16);
+	void (*deactivate)(void *, u16, u16);
+	void (*set_affinity)(void *, u16, u16, u8, u32);
+	void *(*get)(struct irq_remap_table *, int);
+	void (*set_allocated)(struct irq_remap_table *, int);
+	bool (*is_allocated)(struct irq_remap_table *, int);
+	void (*clear_allocated)(struct irq_remap_table *, int);
 };
 
+#ifdef CONFIG_IRQ_REMAP
+extern struct amd_irte_ops irte_32_ops;
+extern struct amd_irte_ops irte_128_ops;
+#endif
+
 #endif /* _ASM_X86_AMD_IOMMU_TYPES_H */
-- 
1.9.1

[PART2 PATCH v6 04/12] iommu/amd: Add support for multiple IRTE formats

[Suravee Suthikulpanit]

This patch enables support for the new 128-bit IOMMU IRTE format,
which can be used for both legacy and vapic interrupt remapping modes.
It replaces the existing operations on IRTE, which can only support
the older 32-bit IRTE format, with calls to the new struct amd_irt_ops.

It also provides helper functions for setting up, accessing, and
updating interrupt remapping table entries in different mode.

Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
---
 drivers/iommu/amd_iommu.c       | 72 +++++++++++++++++++++++++++--------------
 drivers/iommu/amd_iommu_init.c  |  2 ++
 drivers/iommu/amd_iommu_types.h |  1 -
 3 files changed, 50 insertions(+), 25 deletions(-)

diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
index 139ea8b..abfb2b7 100644
--- a/drivers/iommu/amd_iommu.c
+++ b/drivers/iommu/amd_iommu.c
@@ -3714,8 +3714,6 @@ static void set_dte_irq_entry(u16 devid, struct irq_remap_table *table)
 	amd_iommu_dev_table[devid].data[2] = dte;
 }
 
-#define IRTE_ALLOCATED (~1U)
-
 static struct irq_remap_table *get_irq_table(u16 devid, bool ioapic)
 {
 	struct irq_remap_table *table = NULL;
@@ -3761,13 +3759,18 @@ static struct irq_remap_table *get_irq_table(u16 devid, bool ioapic)
 		goto out;
 	}
 
-	memset(table->table, 0, MAX_IRQS_PER_TABLE * sizeof(u32));
+	if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
+		memset(table->table, 0,
+		       MAX_IRQS_PER_TABLE * sizeof(u32));
+	else
+		memset(table->table, 0,
+		       (MAX_IRQS_PER_TABLE * (sizeof(u64) * 2)));
 
 	if (ioapic) {
 		int i;
 
 		for (i = 0; i < 32; ++i)
-			table->table[i] = IRTE_ALLOCATED;
+			iommu->irte_ops->set_allocated(table, i);
 	}
 
 	irq_lookup_table[devid] = table;
@@ -3793,6 +3796,10 @@ static int alloc_irq_index(u16 devid, int count)
 	struct irq_remap_table *table;
 	unsigned long flags;
 	int index, c;
+	struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
+
+	if (!iommu)
+		return -ENODEV;
 
 	table = get_irq_table(devid, false);
 	if (!table)
@@ -3804,14 +3811,14 @@ static int alloc_irq_index(u16 devid, int count)
 	for (c = 0, index = table->min_index;
 	     index < MAX_IRQS_PER_TABLE;
 	     ++index) {
-		if (table->table[index] == 0)
+		if (!iommu->irte_ops->is_allocated(table, index))
 			c += 1;
 		else
 			c = 0;
 
 		if (c == count)	{
 			for (; c != 0; --c)
-				table->table[index - c + 1] = IRTE_ALLOCATED;
+				iommu->irte_ops->set_allocated(table, index - c + 1);
 
 			index -= count - 1;
 			goto out;
@@ -3900,7 +3907,7 @@ static void free_irte(u16 devid, int index)
 		return;
 
 	spin_lock_irqsave(&table->lock, flags);
-	table->table[index] = 0;
+	iommu->irte_ops->clear_allocated(table, index);
 	spin_unlock_irqrestore(&table->lock, flags);
 
 	iommu_flush_irt(iommu, devid);
@@ -3990,6 +3997,7 @@ static void irte_ga_set_affinity(void *entry, u16 devid, u16 index,
 	modify_irte_ga(devid, index, irte);
 }
 
+#define IRTE_ALLOCATED (~1U)
 static void irte_set_allocated(struct irq_remap_table *table, int index)
 {
 	table->table[index] = IRTE_ALLOCATED;
@@ -4119,19 +4127,17 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data,
 {
 	struct irq_2_irte *irte_info = &data->irq_2_irte;
 	struct msi_msg *msg = &data->msi_entry;
-	union irte *irte = &data->irte_entry;
 	struct IO_APIC_route_entry *entry;
+	struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
+
+	if (!iommu)
+		return;
 
 	data->irq_2_irte.devid = devid;
 	data->irq_2_irte.index = index + sub_handle;
-
-	/* Setup IRTE for IOMMU */
-	irte->val = 0;
-	irte->fields.vector      = irq_cfg->vector;
-	irte->fields.int_type    = apic->irq_delivery_mode;
-	irte->fields.destination = irq_cfg->dest_apicid;
-	irte->fields.dm          = apic->irq_dest_mode;
-	irte->fields.valid       = 1;
+	iommu->irte_ops->prepare(data->entry, apic->irq_delivery_mode,
+				 apic->irq_dest_mode, irq_cfg->vector,
+				 irq_cfg->dest_apicid);
 
 	switch (info->type) {
 	case X86_IRQ_ALLOC_TYPE_IOAPIC:
@@ -4187,7 +4193,7 @@ static int irq_remapping_alloc(struct irq_domain *domain, unsigned int virq,
 {
 	struct irq_alloc_info *info = arg;
 	struct irq_data *irq_data;
-	struct amd_ir_data *data;
+	struct amd_ir_data *data = NULL;
 	struct irq_cfg *cfg;
 	int i, ret, devid;
 	int index = -1;
@@ -4239,6 +4245,16 @@ static int irq_remapping_alloc(struct irq_domain *domain, unsigned int virq,
 		if (!data)
 			goto out_free_data;
 
+		if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
+			data->entry = kzalloc(sizeof(union irte), GFP_KERNEL);
+		else
+			data->entry = kzalloc(sizeof(struct irte_ga),
+						     GFP_KERNEL);
+		if (!data->entry) {
+			kfree(data);
+			goto out_free_data;
+		}
+
 		irq_data->hwirq = (devid << 16) + i;
 		irq_data->chip_data = data;
 		irq_data->chip = &amd_ir_chip;
@@ -4275,6 +4291,7 @@ static void irq_remapping_free(struct irq_domain *domain, unsigned int virq,
 			data = irq_data->chip_data;
 			irte_info = &data->irq_2_irte;
 			free_irte(irte_info->devid, irte_info->index);
+			kfree(data->entry);
 			kfree(data);
 		}
 	}
@@ -4286,8 +4303,11 @@ static void irq_remapping_activate(struct irq_domain *domain,
 {
 	struct amd_ir_data *data = irq_data->chip_data;
 	struct irq_2_irte *irte_info = &data->irq_2_irte;
+	struct amd_iommu *iommu = amd_iommu_rlookup_table[irte_info->devid];
 
-	modify_irte(irte_info->devid, irte_info->index, &data->irte_entry);
+	if (iommu)
+		iommu->irte_ops->activate(data->entry, irte_info->devid,
+					  irte_info->index);
 }
 
 static void irq_remapping_deactivate(struct irq_domain *domain,
@@ -4295,10 +4315,11 @@ static void irq_remapping_deactivate(struct irq_domain *domain,
 {
 	struct amd_ir_data *data = irq_data->chip_data;
 	struct irq_2_irte *irte_info = &data->irq_2_irte;
-	union irte entry;
+	struct amd_iommu *iommu = amd_iommu_rlookup_table[irte_info->devid];
 
-	entry.val = 0;
-	modify_irte(irte_info->devid, irte_info->index, &data->irte_entry);
+	if (iommu)
+		iommu->irte_ops->deactivate(data->entry, irte_info->devid,
+					    irte_info->index);
 }
 
 static struct irq_domain_ops amd_ir_domain_ops = {
@@ -4315,8 +4336,12 @@ static int amd_ir_set_affinity(struct irq_data *data,
 	struct irq_2_irte *irte_info = &ir_data->irq_2_irte;
 	struct irq_cfg *cfg = irqd_cfg(data);
 	struct irq_data *parent = data->parent_data;
+	struct amd_iommu *iommu = amd_iommu_rlookup_table[irte_info->devid];
 	int ret;
 
+	if (!iommu)
+		return -ENODEV;
+
 	ret = parent->chip->irq_set_affinity(parent, mask, force);
 	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
 		return ret;
@@ -4325,9 +4350,8 @@ static int amd_ir_set_affinity(struct irq_data *data,
 	 * Atomically updates the IRTE with the new destination, vector
 	 * and flushes the interrupt entry cache.
 	 */
-	ir_data->irte_entry.fields.vector = cfg->vector;
-	ir_data->irte_entry.fields.destination = cfg->dest_apicid;
-	modify_irte(irte_info->devid, irte_info->index, &ir_data->irte_entry);
+	iommu->irte_ops->set_affinity(ir_data->entry, irte_info->devid,
+			    irte_info->index, cfg->vector, cfg->dest_apicid);
 
 	/*
 	 * After this point, all the interrupts will start arriving
diff --git a/drivers/iommu/amd_iommu_init.c b/drivers/iommu/amd_iommu_init.c
index c3afd86..c17febb 100644
--- a/drivers/iommu/amd_iommu_init.c
+++ b/drivers/iommu/amd_iommu_init.c
@@ -1893,8 +1893,10 @@ static void iommu_enable_ga(struct amd_iommu *iommu)
 		/* Fall through */
 	case AMD_IOMMU_GUEST_IR_LEGACY_GA:
 		iommu_feature_enable(iommu, CONTROL_GA_EN);
+		iommu->irte_ops = &irte_128_ops;
 		break;
 	default:
+		iommu->irte_ops = &irte_32_ops;
 		break;
 	}
 #endif
diff --git a/drivers/iommu/amd_iommu_types.h b/drivers/iommu/amd_iommu_types.h
index d57c56d..a591747 100644
--- a/drivers/iommu/amd_iommu_types.h
+++ b/drivers/iommu/amd_iommu_types.h
@@ -781,7 +781,6 @@ struct irq_2_irte {
 
 struct amd_ir_data {
 	struct irq_2_irte irq_2_irte;
-	union irte irte_entry;
 	struct msi_msg msi_entry;
 	void *entry;    /* Pointer to union irte or struct irte_ga */
 };
-- 
1.9.1

[PART2 PATCH v6 05/12] iommu/amd: Detect and initialize guest vAPIC log

[Suravee Suthikulpanit]

This patch adds support to detect and initialize IOMMU Guest vAPIC log
(GALOG). By default, it also enable GALog interrupt to notify IOMMU driver
when GA Log entry is created.

Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
---
 drivers/iommu/amd_iommu_init.c  | 112 +++++++++++++++++++++++++++++++++++++---
 drivers/iommu/amd_iommu_types.h |  28 ++++++++++
 2 files changed, 133 insertions(+), 7 deletions(-)

diff --git a/drivers/iommu/amd_iommu_init.c b/drivers/iommu/amd_iommu_init.c
index c17febb..156ab4b 100644
--- a/drivers/iommu/amd_iommu_init.c
+++ b/drivers/iommu/amd_iommu_init.c
@@ -84,6 +84,7 @@
 #define ACPI_DEVFLAG_LINT1              0x80
 #define ACPI_DEVFLAG_ATSDIS             0x10000000
 
+#define LOOP_TIMEOUT	100000
 /*
  * ACPI table definitions
  *
@@ -388,6 +389,10 @@ static void iommu_disable(struct amd_iommu *iommu)
 	iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
 
+	/* Disable IOMMU GA_LOG */
+	iommu_feature_disable(iommu, CONTROL_GALOG_EN);
+	iommu_feature_disable(iommu, CONTROL_GAINT_EN);
+
 	/* Disable IOMMU hardware itself */
 	iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
 }
@@ -673,6 +678,99 @@ static void __init free_ppr_log(struct amd_iommu *iommu)
 	free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
 }
 
+static void free_ga_log(struct amd_iommu *iommu)
+{
+#ifdef CONFIG_IRQ_REMAP
+	if (iommu->ga_log)
+		free_pages((unsigned long)iommu->ga_log,
+			    get_order(GA_LOG_SIZE));
+	if (iommu->ga_log_tail)
+		free_pages((unsigned long)iommu->ga_log_tail,
+			    get_order(8));
+#endif
+}
+
+static int iommu_ga_log_enable(struct amd_iommu *iommu)
+{
+#ifdef CONFIG_IRQ_REMAP
+	u32 status, i;
+
+	if (!iommu->ga_log)
+		return -EINVAL;
+
+	status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
+
+	/* Check if already running */
+	if (status & (MMIO_STATUS_GALOG_RUN_MASK))
+		return 0;
+
+	iommu_feature_enable(iommu, CONTROL_GAINT_EN);
+	iommu_feature_enable(iommu, CONTROL_GALOG_EN);
+
+	for (i = 0; i < LOOP_TIMEOUT; ++i) {
+		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
+		if (status & (MMIO_STATUS_GALOG_RUN_MASK))
+			break;
+	}
+
+	if (i >= LOOP_TIMEOUT)
+		return -EINVAL;
+#endif /* CONFIG_IRQ_REMAP */
+	return 0;
+}
+
+#ifdef CONFIG_IRQ_REMAP
+static int iommu_init_ga_log(struct amd_iommu *iommu)
+{
+	u64 entry;
+
+	if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
+		return 0;
+
+	iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
+					get_order(GA_LOG_SIZE));
+	if (!iommu->ga_log)
+		goto err_out;
+
+	iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
+					get_order(8));
+	if (!iommu->ga_log_tail)
+		goto err_out;
+
+	entry = (u64)virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
+	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
+		    &entry, sizeof(entry));
+	entry = ((u64)virt_to_phys(iommu->ga_log) & 0xFFFFFFFFFFFFFULL) & ~7ULL;
+	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
+		    &entry, sizeof(entry));
+	writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
+	writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
+
+	return 0;
+err_out:
+	free_ga_log(iommu);
+	return -EINVAL;
+}
+#endif /* CONFIG_IRQ_REMAP */
+
+static int iommu_init_ga(struct amd_iommu *iommu)
+{
+	int ret = 0;
+
+#ifdef CONFIG_IRQ_REMAP
+	/* Note: We have already checked GASup from IVRS table.
+	 *       Now, we need to make sure that GAMSup is set.
+	 */
+	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
+	    !iommu_feature(iommu, FEATURE_GAM_VAPIC))
+		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
+
+	ret = iommu_init_ga_log(iommu);
+#endif /* CONFIG_IRQ_REMAP */
+
+	return ret;
+}
+
 static void iommu_enable_gt(struct amd_iommu *iommu)
 {
 	if (!iommu_feature(iommu, FEATURE_GT))
@@ -1146,6 +1244,7 @@ static void __init free_iommu_one(struct amd_iommu *iommu)
 	free_command_buffer(iommu);
 	free_event_buffer(iommu);
 	free_ppr_log(iommu);
+	free_ga_log(iommu);
 	iommu_unmap_mmio_space(iommu);
 }
 
@@ -1438,6 +1537,7 @@ static int iommu_init_pci(struct amd_iommu *iommu)
 {
 	int cap_ptr = iommu->cap_ptr;
 	u32 range, misc, low, high;
+	int ret;
 
 	iommu->dev = pci_get_bus_and_slot(PCI_BUS_NUM(iommu->devid),
 					  iommu->devid & 0xff);
@@ -1494,13 +1594,9 @@ static int iommu_init_pci(struct amd_iommu *iommu)
 	if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
 		return -ENOMEM;
 
-	/* Note: We have already checked GASup from IVRS table.
-	 *       Now, we need to make sure that GAMSup is set.
-	 */
-	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
-	    !iommu_feature(iommu, FEATURE_GAM_VAPIC))
-		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
-
+	ret = iommu_init_ga(iommu);
+	if (ret)
+		return ret;
 
 	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
 		amd_iommu_np_cache = true;
@@ -1667,6 +1763,8 @@ enable_faults:
 	if (iommu->ppr_log != NULL)
 		iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
 
+	iommu_ga_log_enable(iommu);
+
 	return 0;
 }
 
diff --git a/drivers/iommu/amd_iommu_types.h b/drivers/iommu/amd_iommu_types.h
index a591747..a3b6e22 100644
--- a/drivers/iommu/amd_iommu_types.h
+++ b/drivers/iommu/amd_iommu_types.h
@@ -70,6 +70,8 @@
 #define MMIO_EXCL_LIMIT_OFFSET  0x0028
 #define MMIO_EXT_FEATURES	0x0030
 #define MMIO_PPR_LOG_OFFSET	0x0038
+#define MMIO_GA_LOG_BASE_OFFSET	0x00e0
+#define MMIO_GA_LOG_TAIL_OFFSET	0x00e8
 #define MMIO_CMD_HEAD_OFFSET	0x2000
 #define MMIO_CMD_TAIL_OFFSET	0x2008
 #define MMIO_EVT_HEAD_OFFSET	0x2010
@@ -77,6 +79,8 @@
 #define MMIO_STATUS_OFFSET	0x2020
 #define MMIO_PPR_HEAD_OFFSET	0x2030
 #define MMIO_PPR_TAIL_OFFSET	0x2038
+#define MMIO_GA_HEAD_OFFSET	0x2040
+#define MMIO_GA_TAIL_OFFSET	0x2048
 #define MMIO_CNTR_CONF_OFFSET	0x4000
 #define MMIO_CNTR_REG_OFFSET	0x40000
 #define MMIO_REG_END_OFFSET	0x80000
@@ -112,6 +116,9 @@
 #define MMIO_STATUS_EVT_INT_MASK	(1 << 1)
 #define MMIO_STATUS_COM_WAIT_INT_MASK	(1 << 2)
 #define MMIO_STATUS_PPR_INT_MASK	(1 << 6)
+#define MMIO_STATUS_GALOG_RUN_MASK	(1 << 8)
+#define MMIO_STATUS_GALOG_OVERFLOW_MASK	(1 << 9)
+#define MMIO_STATUS_GALOG_INT_MASK	(1 << 10)
 
 /* event logging constants */
 #define EVENT_ENTRY_SIZE	0x10
@@ -150,6 +157,8 @@
 #define CONTROL_GT_EN           0x10ULL
 #define CONTROL_GA_EN           0x11ULL
 #define CONTROL_GAM_EN          0x19ULL
+#define CONTROL_GALOG_EN        0x1CULL
+#define CONTROL_GAINT_EN        0x1DULL
 
 #define CTRL_INV_TO_MASK	(7 << CONTROL_INV_TIMEOUT)
 #define CTRL_INV_TO_NONE	0
@@ -228,6 +237,19 @@
 
 #define PPR_REQ_FAULT		0x01
 
+/* Constants for GA Log handling */
+#define GA_LOG_ENTRIES		512
+#define GA_LOG_SIZE_SHIFT	56
+#define GA_LOG_SIZE_512		(0x8ULL << GA_LOG_SIZE_SHIFT)
+#define GA_ENTRY_SIZE		8
+#define GA_LOG_SIZE		(GA_ENTRY_SIZE * GA_LOG_ENTRIES)
+
+#define GA_TAG(x)		(u32)(x & 0xffffffffULL)
+#define GA_DEVID(x)		(u16)(((x) >> 32) & 0xffffULL)
+#define GA_REQ_TYPE(x)		(((x) >> 60) & 0xfULL)
+
+#define GA_GUEST_NR		0x1
+
 #define PAGE_MODE_NONE    0x00
 #define PAGE_MODE_1_LEVEL 0x01
 #define PAGE_MODE_2_LEVEL 0x02
@@ -502,6 +524,12 @@ struct amd_iommu {
 	/* Base of the PPR log, if present */
 	u8 *ppr_log;
 
+	/* Base of the GA log, if present */
+	u8 *ga_log;
+
+	/* Tail of the GA log, if present */
+	u8 *ga_log_tail;
+
 	/* true if interrupts for this IOMMU are already enabled */
 	bool int_enabled;
 
-- 
1.9.1

[PART2 PATCH v6 06/12] iommu/amd: Adding GALOG interrupt handler

[Suravee Suthikulpanit]

This patch adds AMD IOMMU guest virtual APIC log (GALOG) handler.
When IOMMU hardware receives an interrupt targeting a blocking vcpu,
it creates an entry in the GALOG, and generates an interrupt to notify
the AMD IOMMU driver.

At this point, the driver processes the log entry, and notify the SVM
driver via the registered iommu_ga_log_notifier function.

Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
---
 drivers/iommu/amd_iommu.c | 73 +++++++++++++++++++++++++++++++++++++++++++++--
 include/linux/amd-iommu.h | 20 +++++++++++--
 2 files changed, 87 insertions(+), 6 deletions(-)

diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
index abfb2b7..fd5ceff 100644
--- a/drivers/iommu/amd_iommu.c
+++ b/drivers/iommu/amd_iommu.c
@@ -741,14 +741,74 @@ static void iommu_poll_ppr_log(struct amd_iommu *iommu)
 	}
 }
 
+#ifdef CONFIG_IRQ_REMAP
+static int (*iommu_ga_log_notifier)(u32);
+
+int amd_iommu_register_ga_log_notifier(int (*notifier)(u32))
+{
+	iommu_ga_log_notifier = notifier;
+
+	return 0;
+}
+EXPORT_SYMBOL(amd_iommu_register_ga_log_notifier);
+
+static void iommu_poll_ga_log(struct amd_iommu *iommu)
+{
+	u32 head, tail, cnt = 0;
+
+	if (iommu->ga_log == NULL)
+		return;
+
+	head = readl(iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
+	tail = readl(iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
+
+	while (head != tail) {
+		volatile u64 *raw;
+		u64 log_entry;
+
+		raw = (u64 *)(iommu->ga_log + head);
+		cnt++;
+
+		/* Avoid memcpy function-call overhead */
+		log_entry = *raw;
+
+		/* Update head pointer of hardware ring-buffer */
+		head = (head + GA_ENTRY_SIZE) % GA_LOG_SIZE;
+		writel(head, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
+
+		/* Handle GA entry */
+		switch (GA_REQ_TYPE(log_entry)) {
+		case GA_GUEST_NR:
+			if (!iommu_ga_log_notifier)
+				break;
+
+			pr_debug("AMD-Vi: %s: devid=%#x, ga_tag=%#x\n",
+				 __func__, GA_DEVID(log_entry),
+				 GA_TAG(log_entry));
+
+			if (iommu_ga_log_notifier(GA_TAG(log_entry)) != 0)
+				pr_err("AMD-Vi: GA log notifier failed.\n");
+			break;
+		default:
+			break;
+		}
+	}
+}
+#endif /* CONFIG_IRQ_REMAP */
+
+#define AMD_IOMMU_INT_MASK	\
+	(MMIO_STATUS_EVT_INT_MASK | \
+	 MMIO_STATUS_PPR_INT_MASK | \
+	 MMIO_STATUS_GALOG_INT_MASK)
+
 irqreturn_t amd_iommu_int_thread(int irq, void *data)
 {
 	struct amd_iommu *iommu = (struct amd_iommu *) data;
 	u32 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
 
-	while (status & (MMIO_STATUS_EVT_INT_MASK | MMIO_STATUS_PPR_INT_MASK)) {
-		/* Enable EVT and PPR interrupts again */
-		writel((MMIO_STATUS_EVT_INT_MASK | MMIO_STATUS_PPR_INT_MASK),
+	while (status & AMD_IOMMU_INT_MASK) {
+		/* Enable EVT and PPR and GA interrupts again */
+		writel(AMD_IOMMU_INT_MASK,
 			iommu->mmio_base + MMIO_STATUS_OFFSET);
 
 		if (status & MMIO_STATUS_EVT_INT_MASK) {
@@ -761,6 +821,13 @@ irqreturn_t amd_iommu_int_thread(int irq, void *data)
 			iommu_poll_ppr_log(iommu);
 		}
 
+#ifdef CONFIG_IRQ_REMAP
+		if (status & MMIO_STATUS_GALOG_INT_MASK) {
+			pr_devel("AMD-Vi: Processing IOMMU GA Log\n");
+			iommu_poll_ga_log(iommu);
+		}
+#endif
+
 		/*
 		 * Hardware bug: ERBT1312
 		 * When re-enabling interrupt (by writing 1
diff --git a/include/linux/amd-iommu.h b/include/linux/amd-iommu.h
index 2b08e79..465d096 100644
--- a/include/linux/amd-iommu.h
+++ b/include/linux/amd-iommu.h
@@ -168,11 +168,25 @@ typedef void (*amd_iommu_invalidate_ctx)(struct pci_dev *pdev, int pasid);
 
 extern int amd_iommu_set_invalidate_ctx_cb(struct pci_dev *pdev,
 					   amd_iommu_invalidate_ctx cb);
-
-#else
+#else /* CONFIG_AMD_IOMMU */
 
 static inline int amd_iommu_detect(void) { return -ENODEV; }
 
-#endif
+#endif /* CONFIG_AMD_IOMMU */
+
+#if defined(CONFIG_AMD_IOMMU) && defined(CONFIG_IRQ_REMAP)
+
+/* IOMMU AVIC Function */
+extern int amd_iommu_register_ga_log_notifier(int (*notifier)(u32));
+
+#else /* defined(CONFIG_AMD_IOMMU) && defined(CONFIG_IRQ_REMAP) */
+
+static inline int
+amd_iommu_register_ga_log_notifier(int (*notifier)(u32))
+{
+	return 0;
+}
+
+#endif /* defined(CONFIG_AMD_IOMMU) && defined(CONFIG_IRQ_REMAP) */
 
 #endif /* _ASM_X86_AMD_IOMMU_H */
-- 
1.9.1

[PART2 PATCH v6 07/12] iommu/amd: Introduce amd_iommu_update_ga()

[Suravee Suthikulpanit]

Introduces a new IOMMU API, amd_iommu_update_ga(), which allows
KVM (SVM) to update existing posted interrupt IOMMU IRTE when
load/unload vcpu.

Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
---
 drivers/iommu/amd_iommu.c       | 40 ++++++++++++++++++++++++++++++++++++++++
 drivers/iommu/amd_iommu_types.h | 11 -----------
 include/linux/amd-iommu.h       | 22 ++++++++++++++++++++++
 3 files changed, 62 insertions(+), 11 deletions(-)

diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
index fd5ceff..8bb85ca 100644
--- a/drivers/iommu/amd_iommu.c
+++ b/drivers/iommu/amd_iommu.c
@@ -4454,4 +4454,44 @@ int amd_iommu_create_irq_domain(struct amd_iommu *iommu)
 
 	return 0;
 }
+
+int amd_iommu_update_ga(u32 cpu, u64 base, bool is_run,
+			struct amd_ir_data *ir_data)
+{
+	unsigned long flags;
+	struct amd_iommu *iommu;
+	struct irq_remap_table *irt;
+	int devid = ir_data->irq_2_irte.devid;
+	struct irte_ga *entry = (struct irte_ga *) ir_data->entry;
+	struct irte_ga *ref = (struct irte_ga *) ir_data->ref;
+
+	if ((!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir)) ||
+	    (!entry->lo.fields_vapic.guest_mode))
+		return 0;
+
+	iommu = amd_iommu_rlookup_table[devid];
+	if (!iommu)
+		return -ENODEV;
+
+	irt = get_irq_table(devid, false);
+	if (!irt)
+		return -ENODEV;
+
+	spin_lock_irqsave(&irt->lock, flags);
+
+	if (ref->lo.fields_vapic.guest_mode) {
+		ref->hi.fields.ga_root_ptr = (base >> 12);
+		if (cpu >= 0)
+			ref->lo.fields_vapic.destination = cpu;
+		ref->lo.fields_vapic.is_run = is_run;
+		barrier();
+	}
+
+	spin_unlock_irqrestore(&irt->lock, flags);
+
+	iommu_flush_irt(iommu, devid);
+	iommu_completion_wait(iommu);
+	return 0;
+}
+EXPORT_SYMBOL(amd_iommu_update_ga);
 #endif
diff --git a/drivers/iommu/amd_iommu_types.h b/drivers/iommu/amd_iommu_types.h
index a3b6e22..71d1d0e 100644
--- a/drivers/iommu/amd_iommu_types.h
+++ b/drivers/iommu/amd_iommu_types.h
@@ -802,17 +802,6 @@ struct irte_ga {
 	union irte_ga_hi hi;
 };
 
-struct irq_2_irte {
-	u16 devid; /* Device ID for IRTE table */
-	u16 index; /* Index into IRTE table*/
-};
-
-struct amd_ir_data {
-	struct irq_2_irte irq_2_irte;
-	struct msi_msg msi_entry;
-	void *entry;    /* Pointer to union irte or struct irte_ga */
-};
-
 struct amd_irte_ops {
 	void (*prepare)(void *, u32, u32, u8, u32);
 	void (*activate)(void *, u16, u16);
diff --git a/include/linux/amd-iommu.h b/include/linux/amd-iommu.h
index 465d096..dfd3bc4 100644
--- a/include/linux/amd-iommu.h
+++ b/include/linux/amd-iommu.h
@@ -21,6 +21,19 @@
 #define _ASM_X86_AMD_IOMMU_H
 
 #include <linux/types.h>
+#include <linux/msi.h>
+
+struct irq_2_irte {
+	u16 devid; /* Device ID for IRTE table */
+	u16 index; /* Index into IRTE table*/
+};
+
+struct amd_ir_data {
+	struct irq_2_irte irq_2_irte;
+	struct msi_msg msi_entry;
+	void *entry;    /* Pointer to union irte or struct irte_ga */
+	void *ref;      /* Pointer to the actual irte */
+};
 
 #ifdef CONFIG_AMD_IOMMU
 
@@ -179,6 +192,9 @@ static inline int amd_iommu_detect(void) { return -ENODEV; }
 /* IOMMU AVIC Function */
 extern int amd_iommu_register_ga_log_notifier(int (*notifier)(u32));
 
+extern int
+amd_iommu_update_ga(u32 cpu, u64 base, bool is_run, struct amd_ir_data *ir_data);
+
 #else /* defined(CONFIG_AMD_IOMMU) && defined(CONFIG_IRQ_REMAP) */
 
 static inline int
@@ -187,6 +203,12 @@ amd_iommu_register_ga_log_notifier(int (*notifier)(u32))
 	return 0;
 }
 
+static inline int
+amd_iommu_update_ga(u32 cpu, u64 base, bool is_run, struct amd_ir_data *ir_data);
+{
+	return 0;
+}
+
 #endif /* defined(CONFIG_AMD_IOMMU) && defined(CONFIG_IRQ_REMAP) */
 
 #endif /* _ASM_X86_AMD_IOMMU_H */
-- 
1.9.1

[PART2 PATCH v6 08/12] iommu/amd: Implements irq_set_vcpu_affinity() hook to setup vapic mode for pass-through devices

[Suravee Suthikulpanit]

This patch implements irq_set_vcpu_affinity() function to set up interrupt
remapping table entry with vapic mode for pass-through devices.

In case requirements for vapic mode are not met, it falls back to set up
the IRTE in legacy mode.

Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
---
 drivers/iommu/amd_iommu.c | 67 ++++++++++++++++++++++++++++++++++++++++++++---
 include/linux/amd-iommu.h | 13 +++++++++
 2 files changed, 76 insertions(+), 4 deletions(-)

diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
index 8bb85ca..65cdfcd 100644
--- a/drivers/iommu/amd_iommu.c
+++ b/drivers/iommu/amd_iommu.c
@@ -3900,7 +3900,8 @@ out:
 	return index;
 }
 
-static int modify_irte_ga(u16 devid, int index, struct irte_ga *irte)
+static int modify_irte_ga(u16 devid, int index, struct irte_ga *irte,
+			  struct amd_ir_data *data)
 {
 	struct irq_remap_table *table;
 	struct amd_iommu *iommu;
@@ -3926,6 +3927,8 @@ static int modify_irte_ga(u16 devid, int index, struct irte_ga *irte)
 	entry->hi.fields.ga_root_ptr = tmp.hi.fields.ga_root_ptr;
 	entry->lo.val = irte->lo.val;
 	entry->lo.fields_remap.valid = 1;
+	if (data)
+		data->ref = entry;
 
 	spin_unlock_irqrestore(&table->lock, flags);
 
@@ -4024,7 +4027,7 @@ static void irte_ga_activate(void *entry, u16 devid, u16 index)
 	struct irte_ga *irte = (struct irte_ga *) entry;
 
 	irte->lo.fields_remap.valid = 1;
-	modify_irte_ga(devid, index, irte);
+	modify_irte_ga(devid, index, irte, NULL);
 }
 
 static void irte_deactivate(void *entry, u16 devid, u16 index)
@@ -4040,7 +4043,7 @@ static void irte_ga_deactivate(void *entry, u16 devid, u16 index)
 	struct irte_ga *irte = (struct irte_ga *) entry;
 
 	irte->lo.fields_remap.valid = 0;
-	modify_irte_ga(devid, index, irte);
+	modify_irte_ga(devid, index, irte, NULL);
 }
 
 static void irte_set_affinity(void *entry, u16 devid, u16 index,
@@ -4061,7 +4064,7 @@ static void irte_ga_set_affinity(void *entry, u16 devid, u16 index,
 	irte->hi.fields.vector = vector;
 	irte->lo.fields_remap.destination = dest_apicid;
 	irte->lo.fields_remap.guest_mode = 0;
-	modify_irte_ga(devid, index, irte);
+	modify_irte_ga(devid, index, irte, NULL);
 }
 
 #define IRTE_ALLOCATED (~1U)
@@ -4396,6 +4399,61 @@ static struct irq_domain_ops amd_ir_domain_ops = {
 	.deactivate = irq_remapping_deactivate,
 };
 
+static int amd_ir_set_vcpu_affinity(struct irq_data *data, void *vcpu_info)
+{
+	struct amd_iommu *iommu;
+	struct amd_iommu_pi_data *pi_data = vcpu_info;
+	struct vcpu_data *vcpu_pi_info = pi_data->vcpu_data;
+	struct amd_ir_data *ir_data = data->chip_data;
+	struct irte_ga *irte = (struct irte_ga *) ir_data->entry;
+	struct irq_2_irte *irte_info = &ir_data->irq_2_irte;
+
+	pi_data->ir_data = ir_data;
+
+	/* Note:
+	 * SVM tries to set up for VAPIC mode, but we are in
+	 * legacy mode. So, we force legacy mode instead.
+	 */
+	if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir)) {
+		pr_debug("AMD-Vi: %s: Fall back to using intr legacy remap\n",
+			 __func__);
+		pi_data->is_guest_mode = false;
+	}
+
+	iommu = amd_iommu_rlookup_table[irte_info->devid];
+	if (iommu == NULL)
+		return -EINVAL;
+
+	if (pi_data->is_guest_mode) {
+		/* Setting */
+		irte->hi.fields.vector = vcpu_pi_info->vector;
+		irte->lo.fields_vapic.guest_mode = 1;
+		irte->lo.fields_vapic.ga_tag = pi_data->ga_tag;
+
+		ir_data->cached_ga_tag = pi_data->ga_tag;
+	} else {
+		/* Un-Setting */
+		struct irq_cfg *cfg = irqd_cfg(data);
+
+		irte->hi.val = 0;
+		irte->lo.val = 0;
+		irte->hi.fields.vector = cfg->vector;
+		irte->lo.fields_remap.guest_mode = 0;
+		irte->lo.fields_remap.destination = cfg->dest_apicid;
+		irte->lo.fields_remap.int_type = apic->irq_delivery_mode;
+		irte->lo.fields_remap.dm = apic->irq_dest_mode;
+
+		/*
+		 * This communicates the ga_tag back to the caller
+		 * so that it can do all the necessary clean up.
+		 */
+		pi_data->ga_tag = ir_data->cached_ga_tag;
+		ir_data->cached_ga_tag = 0;
+	}
+
+	return modify_irte_ga(irte_info->devid, irte_info->index, irte, ir_data);
+}
+
 static int amd_ir_set_affinity(struct irq_data *data,
 			       const struct cpumask *mask, bool force)
 {
@@ -4440,6 +4498,7 @@ static void ir_compose_msi_msg(struct irq_data *irq_data, struct msi_msg *msg)
 static struct irq_chip amd_ir_chip = {
 	.irq_ack = ir_ack_apic_edge,
 	.irq_set_affinity = amd_ir_set_affinity,
+	.irq_set_vcpu_affinity = amd_ir_set_vcpu_affinity,
 	.irq_compose_msi_msg = ir_compose_msi_msg,
 };
 
diff --git a/include/linux/amd-iommu.h b/include/linux/amd-iommu.h
index dfd3bc4..7b2e802 100644
--- a/include/linux/amd-iommu.h
+++ b/include/linux/amd-iommu.h
@@ -29,12 +29,25 @@ struct irq_2_irte {
 };
 
 struct amd_ir_data {
+	u32 cached_ga_tag;
 	struct irq_2_irte irq_2_irte;
 	struct msi_msg msi_entry;
 	void *entry;    /* Pointer to union irte or struct irte_ga */
 	void *ref;      /* Pointer to the actual irte */
 };
 
+/*
+ * This is mainly used to communicate information back-and-forth
+ * between SVM and IOMMU for setting up and tearing down posted
+ * interrupt
+ */
+struct amd_iommu_pi_data {
+	u32 ga_tag;
+	bool is_guest_mode;
+	struct vcpu_data *vcpu_data;
+	struct amd_ir_data *ir_data;
+};
+
 #ifdef CONFIG_AMD_IOMMU
 
 struct task_struct;
-- 
1.9.1

[PART2 PATCH v6 09/12] iommu/amd: Enable vAPIC interrupt remapping mode by default

[Suravee Suthikulpanit]

Introduce struct iommu_dev_data.use_vapic flag, which IOMMU driver
uses to determine if it should enable vAPIC support, by setting
the ga_mode bit in the device's interrupt remapping table entry.

Currently, it is enabled for all pass-through device if vAPIC mode
is enabled.

Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
---
 drivers/iommu/amd_iommu.c       | 44 +++++++++++++++++++++++++++++++++--------
 drivers/iommu/amd_iommu_init.c  | 12 ++++++++++-
 drivers/iommu/amd_iommu_types.h |  2 +-
 3 files changed, 48 insertions(+), 10 deletions(-)

diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
index 65cdfcd..9cfac8b 100644
--- a/drivers/iommu/amd_iommu.c
+++ b/drivers/iommu/amd_iommu.c
@@ -105,6 +105,7 @@ struct iommu_dev_data {
 	bool pri_tlp;			  /* PASID TLB required for
 					     PPR completions */
 	u32 errata;			  /* Bitmap for errata to apply */
+	bool use_vapic;			  /* Enable device to use vapic mode */
 };
 
 /*
@@ -3211,6 +3212,12 @@ static void amd_iommu_detach_device(struct iommu_domain *dom,
 	if (!iommu)
 		return;
 
+#ifdef CONFIG_IRQ_REMAP
+	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
+	    (dom->type == IOMMU_DOMAIN_UNMANAGED))
+		dev_data->use_vapic = 0;
+#endif
+
 	iommu_completion_wait(iommu);
 }
 
@@ -3236,6 +3243,15 @@ static int amd_iommu_attach_device(struct iommu_domain *dom,
 
 	ret = attach_device(dev, domain);
 
+#ifdef CONFIG_IRQ_REMAP
+	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir)) {
+		if (dom->type == IOMMU_DOMAIN_UNMANAGED)
+			dev_data->use_vapic = 1;
+		else
+			dev_data->use_vapic = 0;
+	}
+#endif
+
 	iommu_completion_wait(iommu);
 
 	return ret;
@@ -3986,7 +4002,7 @@ static void free_irte(u16 devid, int index)
 
 static void irte_prepare(void *entry,
 			 u32 delivery_mode, u32 dest_mode,
-			 u8 vector, u32 dest_apicid)
+			 u8 vector, u32 dest_apicid, int devid)
 {
 	union irte *irte = (union irte *) entry;
 
@@ -4000,13 +4016,14 @@ static void irte_prepare(void *entry,
 
 static void irte_ga_prepare(void *entry,
 			    u32 delivery_mode, u32 dest_mode,
-			    u8 vector, u32 dest_apicid)
+			    u8 vector, u32 dest_apicid, int devid)
 {
 	struct irte_ga *irte = (struct irte_ga *) entry;
+	struct iommu_dev_data *dev_data = search_dev_data(devid);
 
 	irte->lo.val                      = 0;
 	irte->hi.val                      = 0;
-	irte->lo.fields_remap.guest_mode  = 0;
+	irte->lo.fields_remap.guest_mode  = dev_data ? dev_data->use_vapic : 0;
 	irte->lo.fields_remap.int_type    = delivery_mode;
 	irte->lo.fields_remap.dm          = dest_mode;
 	irte->hi.fields.vector            = vector;
@@ -4060,11 +4077,14 @@ static void irte_ga_set_affinity(void *entry, u16 devid, u16 index,
 				 u8 vector, u32 dest_apicid)
 {
 	struct irte_ga *irte = (struct irte_ga *) entry;
+	struct iommu_dev_data *dev_data = search_dev_data(devid);
 
-	irte->hi.fields.vector = vector;
-	irte->lo.fields_remap.destination = dest_apicid;
-	irte->lo.fields_remap.guest_mode = 0;
-	modify_irte_ga(devid, index, irte, NULL);
+	if (!dev_data || !dev_data->use_vapic) {
+		irte->hi.fields.vector = vector;
+		irte->lo.fields_remap.destination = dest_apicid;
+		irte->lo.fields_remap.guest_mode = 0;
+		modify_irte_ga(devid, index, irte, NULL);
+	}
 }
 
 #define IRTE_ALLOCATED (~1U)
@@ -4207,7 +4227,7 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data,
 	data->irq_2_irte.index = index + sub_handle;
 	iommu->irte_ops->prepare(data->entry, apic->irq_delivery_mode,
 				 apic->irq_dest_mode, irq_cfg->vector,
-				 irq_cfg->dest_apicid);
+				 irq_cfg->dest_apicid, devid);
 
 	switch (info->type) {
 	case X86_IRQ_ALLOC_TYPE_IOAPIC:
@@ -4407,6 +4427,14 @@ static int amd_ir_set_vcpu_affinity(struct irq_data *data, void *vcpu_info)
 	struct amd_ir_data *ir_data = data->chip_data;
 	struct irte_ga *irte = (struct irte_ga *) ir_data->entry;
 	struct irq_2_irte *irte_info = &ir_data->irq_2_irte;
+	struct iommu_dev_data *dev_data = search_dev_data(irte_info->devid);
+
+	/* Note:
+	 * This device has never been set up for guest mode.
+	 * we should not modify the IRTE
+	 */
+	if (!dev_data || !dev_data->use_vapic)
+		return 0;
 
 	pi_data->ir_data = ir_data;
 
diff --git a/drivers/iommu/amd_iommu_init.c b/drivers/iommu/amd_iommu_init.c
index 156ab4b..cd17136 100644
--- a/drivers/iommu/amd_iommu_init.c
+++ b/drivers/iommu/amd_iommu_init.c
@@ -146,7 +146,7 @@ struct ivmd_header {
 bool amd_iommu_dump;
 bool amd_iommu_irq_remap __read_mostly;
 
-int amd_iommu_guest_ir;
+int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
 
 static bool amd_iommu_detected;
 static bool __initdata amd_iommu_disabled;
@@ -2019,6 +2019,11 @@ static void early_enable_iommus(void)
 		iommu_enable(iommu);
 		iommu_flush_all_caches(iommu);
 	}
+
+#ifdef CONFIG_IRQ_REMAP
+	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
+		amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
+#endif
 }
 
 static void enable_iommus_v2(void)
@@ -2044,6 +2049,11 @@ static void disable_iommus(void)
 
 	for_each_iommu(iommu)
 		iommu_disable(iommu);
+
+#ifdef CONFIG_IRQ_REMAP
+	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
+		amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
+#endif
 }
 
 /*
diff --git a/drivers/iommu/amd_iommu_types.h b/drivers/iommu/amd_iommu_types.h
index 71d1d0e..596e43f 100644
--- a/drivers/iommu/amd_iommu_types.h
+++ b/drivers/iommu/amd_iommu_types.h
@@ -803,7 +803,7 @@ struct irte_ga {
 };
 
 struct amd_irte_ops {
-	void (*prepare)(void *, u32, u32, u8, u32);
+	void (*prepare)(void *, u32, u32, u8, u32, int);
 	void (*activate)(void *, u16, u16);
 	void (*deactivate)(void *, u16, u16);
 	void (*set_affinity)(void *, u16, u16, u8, u32);
-- 
1.9.1

[PART2 PATCH v6 10/12] svm: Introduces AVIC per-VM ID

[Suravee Suthikulpanit]

Introduces per-VM AVIC ID and helper functions to manage the IDs.
Currently, the ID will be used to implement 32-bit AVIC IOMMU GA tag.

The ID is 24-bit one-based indexing value, and is managed via helper
functions to get the next ID, or to free an ID once a VM is destroyed.
There should be no ID conflict for any active VMs.

Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
---
 arch/x86/include/asm/kvm_host.h |  1 +
 arch/x86/kvm/svm.c              | 51 +++++++++++++++++++++++++++++++++++++++++
 2 files changed, 52 insertions(+)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 69e62862..16b4d1d 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -776,6 +776,7 @@ struct kvm_arch {
 	bool disabled_lapic_found;
 
 	/* Struct members for AVIC */
+	u32 avic_vm_id;
 	u32 ldr_mode;
 	struct page *avic_logical_id_table_page;
 	struct page *avic_physical_id_table_page;
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index 16ef31b..88acb41 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -96,6 +96,19 @@ MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
 #define AVIC_UNACCEL_ACCESS_OFFSET_MASK		0xFF0
 #define AVIC_UNACCEL_ACCESS_VECTOR_MASK		0xFFFFFFFF
 
+/* AVIC GATAG is encoded using VM and VCPU IDs */
+#define AVIC_VCPU_ID_BITS		8
+#define AVIC_VCPU_ID_MASK		((1 << AVIC_VCPU_ID_BITS) - 1)
+
+#define AVIC_VM_ID_BITS			24
+#define AVIC_VM_ID_NR			(1 << AVIC_VM_ID_BITS)
+#define AVIC_VM_ID_MASK			((1 << AVIC_VM_ID_BITS) - 1)
+
+#define AVIC_GATAG(x, y)		(((x & AVIC_VM_ID_MASK) << AVIC_VCPU_ID_BITS) | \
+						(y & AVIC_VCPU_ID_MASK))
+#define AVIC_GATAG_TO_VMID(x)		((x >> AVIC_VCPU_ID_BITS) & AVIC_VM_ID_MASK)
+#define AVIC_GATAG_TO_VCPUID(x)		(x & AVIC_VM_ID_BITS)
+
 static bool erratum_383_found __read_mostly;
 
 static const u32 host_save_user_msrs[] = {
@@ -242,6 +255,10 @@ static int avic;
 module_param(avic, int, S_IRUGO);
 #endif
 
+/* AVIC VM ID bit masks and lock */
+static DECLARE_BITMAP(avic_vm_id_bitmap, AVIC_VM_ID_NR);
+static DEFINE_SPINLOCK(avic_vm_id_lock);
+
 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
 static void svm_complete_interrupts(struct vcpu_svm *svm);
@@ -1280,10 +1297,40 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 	return 0;
 }
 
+static inline int avic_get_next_vm_id(void)
+{
+	int id;
+
+	spin_lock(&avic_vm_id_lock);
+
+	/* AVIC VM ID is one-based. */
+	id = find_next_zero_bit(avic_vm_id_bitmap, AVIC_VM_ID_NR, 1);
+	if (id <= AVIC_VM_ID_MASK)
+		__set_bit(id, avic_vm_id_bitmap);
+	else
+		id = -EAGAIN;
+
+	spin_unlock(&avic_vm_id_lock);
+	return id;
+}
+
+static inline int avic_free_vm_id(int id)
+{
+	if (id <= 0 || id > AVIC_VM_ID_MASK)
+		return -EINVAL;
+
+	spin_lock(&avic_vm_id_lock);
+	__clear_bit(id, avic_vm_id_bitmap);
+	spin_unlock(&avic_vm_id_lock);
+	return 0;
+}
+
 static void avic_vm_destroy(struct kvm *kvm)
 {
 	struct kvm_arch *vm_data = &kvm->arch;
 
+	avic_free_vm_id(vm_data->avic_vm_id);
+
 	if (vm_data->avic_logical_id_table_page)
 		__free_page(vm_data->avic_logical_id_table_page);
 	if (vm_data->avic_physical_id_table_page)
@@ -1300,6 +1347,10 @@ static int avic_vm_init(struct kvm *kvm)
 	if (!avic)
 		return 0;
 
+	vm_data->avic_vm_id = avic_get_next_vm_id();
+	if (vm_data->avic_vm_id < 0)
+		return vm_data->avic_vm_id;
+
 	/* Allocating physical APIC ID table (4KB) */
 	p_page = alloc_page(GFP_KERNEL);
 	if (!p_page)
-- 
1.9.1

[PART2 PATCH v6 11/12] svm: Introduce AMD IOMMU avic_ga_log_notifier

[Suravee Suthikulpanit]

This patch introduces avic_ga_log_notifier, which will be called
by IOMMU driver whenever it handles the Guest vAPIC (GA) log entry.

Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
---
 arch/x86/include/asm/kvm_host.h |  1 +
 arch/x86/kvm/svm.c              | 73 +++++++++++++++++++++++++++++++++++++++--
 2 files changed, 72 insertions(+), 2 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 16b4d1d..a9466ad 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -780,6 +780,7 @@ struct kvm_arch {
 	u32 ldr_mode;
 	struct page *avic_logical_id_table_page;
 	struct page *avic_physical_id_table_page;
+	struct hlist_node hnode;
 };
 
 struct kvm_vm_stat {
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index 88acb41..c060e05 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -34,6 +34,8 @@
 #include <linux/sched.h>
 #include <linux/trace_events.h>
 #include <linux/slab.h>
+#include <linux/amd-iommu.h>
+#include <linux/hashtable.h>
 
 #include <asm/apic.h>
 #include <asm/perf_event.h>
@@ -945,6 +947,58 @@ static void svm_disable_lbrv(struct vcpu_svm *svm)
 	set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 }
 
+/* Note:
+ * This hash table is used to map VM_ID to a struct kvm_arch,
+ * when handling AMD IOMMU GALOG notification to schedule in
+ * a particular vCPU.
+ */
+#define SVM_VM_DATA_HASH_BITS	8
+DECLARE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS);
+static spinlock_t svm_vm_data_hash_lock;
+
+/* Note:
+ * This function is called from IOMMU driver to notify
+ * SVM to schedule in a particular vCPU of a particular VM.
+ */
+static int avic_ga_log_notifier(u32 ga_tag)
+{
+	unsigned long flags;
+	struct kvm_arch *ka = NULL;
+	struct kvm_vcpu *vcpu = NULL;
+	struct vcpu_svm *svm = NULL;
+	u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag);
+	u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag);
+
+	pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id);
+
+	spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
+	hash_for_each_possible(svm_vm_data_hash, ka, hnode, vm_id) {
+		struct kvm *kvm = container_of(ka, struct kvm, arch);
+		struct kvm_arch *vm_data = &kvm->arch;
+
+		if (vm_data->avic_vm_id != vm_id)
+			continue;
+		vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
+		break;
+	}
+	spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
+
+	if (!vcpu)
+		return 0;
+
+	svm = to_svm(vcpu);
+
+	/* Note:
+	 * At this point, the IOMMU should have already set the pending
+	 * bit in the vAPIC backing page. So, we just need to schedule
+	 * in the vcpu.
+	 */
+	if (vcpu->mode == OUTSIDE_GUEST_MODE)
+		kvm_vcpu_wake_up(vcpu);
+
+	return 0;
+}
+
 static __init int svm_hardware_setup(void)
 {
 	int cpu;
@@ -1003,10 +1057,15 @@ static __init int svm_hardware_setup(void)
 	if (avic) {
 		if (!npt_enabled ||
 		    !boot_cpu_has(X86_FEATURE_AVIC) ||
-		    !IS_ENABLED(CONFIG_X86_LOCAL_APIC))
+		    !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
 			avic = false;
-		else
+		} else {
 			pr_info("AVIC enabled\n");
+
+			hash_init(svm_vm_data_hash);
+			spin_lock_init(&svm_vm_data_hash_lock);
+			amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
+		}
 	}
 
 	return 0;
@@ -1327,6 +1386,7 @@ static inline int avic_free_vm_id(int id)
 
 static void avic_vm_destroy(struct kvm *kvm)
 {
+	unsigned long flags;
 	struct kvm_arch *vm_data = &kvm->arch;
 
 	avic_free_vm_id(vm_data->avic_vm_id);
@@ -1335,10 +1395,15 @@ static void avic_vm_destroy(struct kvm *kvm)
 		__free_page(vm_data->avic_logical_id_table_page);
 	if (vm_data->avic_physical_id_table_page)
 		__free_page(vm_data->avic_physical_id_table_page);
+
+	spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
+	hash_del(&vm_data->hnode);
+	spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
 }
 
 static int avic_vm_init(struct kvm *kvm)
 {
+	unsigned long flags;
 	int err = -ENOMEM;
 	struct kvm_arch *vm_data = &kvm->arch;
 	struct page *p_page;
@@ -1367,6 +1432,10 @@ static int avic_vm_init(struct kvm *kvm)
 	vm_data->avic_logical_id_table_page = l_page;
 	clear_page(page_address(l_page));
 
+	spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
+	hash_add(svm_vm_data_hash, &vm_data->hnode, vm_data->avic_vm_id);
+	spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
+
 	return 0;
 
 free_avic:
-- 
1.9.1

[PART2 PATCH v6 12/12] svm: Implements update_pi_irte hook to setup posted interrupt

[Suravee Suthikulpanit]

This patch implements update_pi_irte function hook to allow SVM
communicate to IOMMU driver regarding how to set up IRTE for handling
posted interrupt.

In case AVIC is enabled, during vcpu_load/unload, SVM needs to update
IOMMU IRTE with appropriate host physical APIC ID. Also, when
vcpu_blocking/unblocking, SVM needs to update the is-running bit in
the IOMMU IRTE. Both are achieved via calling amd_iommu_update_ga().

However, if GA mode is not enabled for the pass-through device,
IOMMU driver will simply just return when calling amd_iommu_update_ga.

Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
---
 arch/x86/kvm/svm.c        | 247 ++++++++++++++++++++++++++++++++++++++++++----
 include/linux/amd-iommu.h |   1 +
 2 files changed, 227 insertions(+), 21 deletions(-)

diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index c060e05..303007a 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -43,6 +43,7 @@
 #include <asm/desc.h>
 #include <asm/debugreg.h>
 #include <asm/kvm_para.h>
+#include <asm/irq_remapping.h>
 
 #include <asm/virtext.h>
 #include "trace.h"
@@ -200,6 +201,16 @@ struct vcpu_svm {
 	struct page *avic_backing_page;
 	u64 *avic_physical_id_cache;
 	bool avic_is_running;
+
+	/*
+	 * Per-vcpu list of struct amd_ir_data:
+	 * This is used mainly to track interrupt remapping table entry (IRTE)
+	 * to be updated when the vcpu affinity changes. This avoid the need
+	 * for scanning for IRTE and try to match ga_tag in the IOMMU driver
+	 * (or using hashtable).
+	 */
+	struct list_head ir_list;
+	spinlock_t ir_list_lock;
 };
 
 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK	(0xFF)
@@ -1443,31 +1454,29 @@ free_avic:
 	return err;
 }
 
-/**
- * This function is called during VCPU halt/unhalt.
- */
-static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run)
+static inline int
+avic_update_iommu(struct kvm_vcpu *vcpu, int cpu, phys_addr_t pa, bool r)
 {
-	u64 entry;
-	int h_physical_id = kvm_cpu_get_apicid(vcpu->cpu);
+	int ret;
+	unsigned long flags;
 	struct vcpu_svm *svm = to_svm(vcpu);
+	struct amd_ir_data *ir;
 
-	if (!kvm_vcpu_apicv_active(vcpu))
-		return;
-
-	svm->avic_is_running = is_run;
-
-	/* ID = 0xff (broadcast), ID > 0xff (reserved) */
-	if (WARN_ON(h_physical_id >= AVIC_MAX_PHYSICAL_ID_COUNT))
-		return;
-
-	entry = READ_ONCE(*(svm->avic_physical_id_cache));
-	WARN_ON(is_run == !!(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK));
+	if (!kvm_arch_has_assigned_device(vcpu->kvm))
+		return 0;
 
-	entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
-	if (is_run)
-		entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
-	WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
+	/*
+	 * Here, we go through the per-vcpu ir_list to update all existing
+	 * interrupt remapping table entry targeting this vcpu.
+	 */
+	spin_lock_irqsave(&svm->ir_list_lock, flags);
+	list_for_each_entry(ir, &svm->ir_list, node) {
+		ret = amd_iommu_update_ga(cpu, (pa & AVIC_HPA_MASK), r, ir);
+		if (ret)
+			break;
+	}
+	spin_unlock_irqrestore(&svm->ir_list_lock, flags);
+	return ret;
 }
 
 static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
@@ -1494,6 +1503,9 @@ static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 		entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
 
 	WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
+	avic_update_iommu(vcpu, h_physical_id,
+			  page_to_phys(svm->avic_backing_page),
+			  svm->avic_is_running);
 }
 
 static void avic_vcpu_put(struct kvm_vcpu *vcpu)
@@ -1505,10 +1517,28 @@ static void avic_vcpu_put(struct kvm_vcpu *vcpu)
 		return;
 
 	entry = READ_ONCE(*(svm->avic_physical_id_cache));
+	if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK)
+		avic_update_iommu(vcpu, -1,
+				  page_to_phys(svm->avic_backing_page), 0);
+
 	entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
 	WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
 }
 
+/**
+ * This function is called during VCPU halt/unhalt.
+ */
+static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run)
+{
+	struct vcpu_svm *svm = to_svm(vcpu);
+
+	svm->avic_is_running = is_run;
+	if (is_run)
+		avic_vcpu_load(vcpu, vcpu->cpu);
+	else
+		avic_vcpu_put(vcpu);
+}
+
 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
@@ -1570,6 +1600,9 @@ static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
 		err = avic_init_backing_page(&svm->vcpu);
 		if (err)
 			goto free_page4;
+
+		INIT_LIST_HEAD(&svm->ir_list);
+		spin_lock_init(&svm->ir_list_lock);
 	}
 
 	/* We initialize this flag to true to make sure that the is_running
@@ -4366,6 +4399,177 @@ static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec)
 		kvm_vcpu_wake_up(vcpu);
 }
 
+static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_ir_data *ir)
+{
+	bool found = false;
+	unsigned long flags;
+	struct amd_ir_data *cur;
+
+	/**
+	 * In some cases, the existing irte is updaed and re-set,
+	 * so we need to check here if it's already been * added
+	 * to the ir_list.
+	 */
+	spin_lock_irqsave(&svm->ir_list_lock, flags);
+	list_for_each_entry(cur, &svm->ir_list, node) {
+		if (cur != ir)
+			continue;
+		found = true;
+		break;
+	}
+	spin_unlock_irqrestore(&svm->ir_list_lock, flags);
+
+	if (found)
+		return 0;
+
+	spin_lock_irqsave(&svm->ir_list_lock, flags);
+	list_add(&ir->node, &svm->ir_list);
+	spin_unlock_irqrestore(&svm->ir_list_lock, flags);
+	return 0;
+}
+
+static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_ir_data *ir)
+{
+	unsigned long flags;
+	struct amd_ir_data *cur;
+
+	spin_lock_irqsave(&svm->ir_list_lock, flags);
+	list_for_each_entry(cur, &svm->ir_list, node) {
+		if (cur != ir)
+			continue;
+		list_del(&cur->node);
+		break;
+	}
+	spin_unlock_irqrestore(&svm->ir_list_lock, flags);
+}
+
+/*
+ * svm_update_pi_irte - set IRTE for Posted-Interrupts
+ *
+ * @kvm: kvm
+ * @host_irq: host irq of the interrupt
+ * @guest_irq: gsi of the interrupt
+ * @set: set or unset PI
+ * returns 0 on success, < 0 on failure
+ */
+static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
+			      uint32_t guest_irq, bool set)
+{
+	struct kvm_kernel_irq_routing_entry *e;
+	struct kvm_irq_routing_table *irq_rt;
+	int idx, ret = -EINVAL;
+
+	if (!kvm_arch_has_assigned_device(kvm) ||
+	    !irq_remapping_cap(IRQ_POSTING_CAP))
+		return 0;
+
+	pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
+		 __func__, host_irq, guest_irq, set);
+
+	idx = srcu_read_lock(&kvm->irq_srcu);
+	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
+	WARN_ON(guest_irq >= irq_rt->nr_rt_entries);
+
+	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
+		struct kvm_lapic_irq irq;
+		struct vcpu_data vcpu_info;
+		struct kvm_vcpu *vcpu = NULL;
+		struct vcpu_svm *svm = NULL;
+
+		if (e->type != KVM_IRQ_ROUTING_MSI)
+			continue;
+
+		/**
+		 * Note:
+		 * The HW cannot support posting multicast/broadcast
+		 * interrupts to a vCPU. So, we still use interrupt
+		 * remapping for these kind of interrupts.
+		 *
+		 * For lowest-priority interrupts, we only support
+		 * those with single CPU as the destination, e.g. user
+		 * configures the interrupts via /proc/irq or uses
+		 * irqbalance to make the interrupts single-CPU.
+		 */
+		kvm_set_msi_irq(e, &irq);
+		if (kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
+			svm = to_svm(vcpu);
+			vcpu_info.pi_desc_addr = page_to_phys(svm->avic_backing_page);
+			vcpu_info.vector = irq.vector;
+
+			pr_debug("SVM: %s: use GA mode for irq %u\n", __func__,
+				 irq.vector);
+		} else {
+			set = false;
+
+			pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
+				 __func__, irq.vector);
+		}
+
+		trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
+					 vcpu_info.vector,
+					 vcpu_info.pi_desc_addr, set);
+
+		/**
+		 * When AVIC is disabled, we fall-back to setup
+		 * IRTE w/ legacy mode
+		 */
+		if (set && kvm_vcpu_apicv_active(&svm->vcpu)) {
+			struct amd_iommu_pi_data pi;
+
+			/* Try to enable guest_mode in IRTE */
+			pi.ga_tag = AVIC_GATAG(kvm->arch.avic_vm_id,
+						     vcpu->vcpu_id);
+			pi.is_guest_mode = true;
+			pi.vcpu_data = &vcpu_info;
+			ret = irq_set_vcpu_affinity(host_irq, &pi);
+
+			/**
+			 * We save the pointer to pi in the struct
+			 * vcpu_svm so that we can reference to them directly
+			 * when we update vcpu scheduling information in IOMMU
+			 * irte.
+			 */
+			if (!ret && pi.is_guest_mode)
+				svm_ir_list_add(svm, pi.ir_data);
+		} else {
+			/* Use legacy mode in IRTE */
+			struct amd_iommu_pi_data pi;
+
+			/**
+			 * Here, pi is used to:
+			 * - Tell IOMMU to use legacy mode for this interrupt.
+			 * - Retrieve ga_tag of prior interrupt remapping data.
+			 */
+			pi.is_guest_mode = false;
+			ret = irq_set_vcpu_affinity(host_irq, &pi);
+
+			/**
+			 * We need to check if the interrupt was previously
+			 * setup with the guest_mode by checking if the ga_tag
+			 * was cached. If so, we need to clean up the per-vcpu
+			 * ir_list.
+			 */
+			if (!ret && pi.ga_tag) {
+				struct kvm_vcpu *vcpu = kvm_get_vcpu_by_id(kvm,
+						AVIC_GATAG_TO_VCPUID(pi.ga_tag));
+
+				if (vcpu)
+					svm_ir_list_del(to_svm(vcpu), pi.ir_data);
+			}
+		}
+
+		if (ret < 0) {
+			pr_err("%s: failed to update PI IRTE\n", __func__);
+			goto out;
+		}
+	}
+
+	ret = 0;
+out:
+	srcu_read_unlock(&kvm->irq_srcu, idx);
+	return ret;
+}
+
 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
@@ -5192,6 +5396,7 @@ static struct kvm_x86_ops svm_x86_ops = {
 
 	.pmu_ops = &amd_pmu_ops,
 	.deliver_posted_interrupt = svm_deliver_avic_intr,
+	.update_pi_irte = svm_update_pi_irte,
 };
 
 static int __init svm_init(void)
diff --git a/include/linux/amd-iommu.h b/include/linux/amd-iommu.h
index 7b2e802..cd10393 100644
--- a/include/linux/amd-iommu.h
+++ b/include/linux/amd-iommu.h
@@ -34,6 +34,7 @@ struct amd_ir_data {
 	struct msi_msg msi_entry;
 	void *entry;    /* Pointer to union irte or struct irte_ga */
 	void *ref;      /* Pointer to the actual irte */
+	struct list_head node;	/* Used by SVM for per-vcpu ir_list */
 };
 
 /*
-- 
1.9.1

Re: [PART2 PATCH v6 10/12] svm: Introduces AVIC per-VM ID

[Radim Krčmář]

2016-08-18 14:42-0500, Suravee Suthikulpanit:
> Introduces per-VM AVIC ID and helper functions to manage the IDs.
> Currently, the ID will be used to implement 32-bit AVIC IOMMU GA tag.
> 
> The ID is 24-bit one-based indexing value, and is managed via helper
> functions to get the next ID, or to free an ID once a VM is destroyed.
> There should be no ID conflict for any active VMs.
> 
> Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
> ---

Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>

Re: [PART2 PATCH v6 12/12] svm: Implements update_pi_irte hook to setup posted interrupt

[Radim Krčmář]

2016-08-18 14:42-0500, Suravee Suthikulpanit:
> This patch implements update_pi_irte function hook to allow SVM
> communicate to IOMMU driver regarding how to set up IRTE for handling
> posted interrupt.
> 
> In case AVIC is enabled, during vcpu_load/unload, SVM needs to update
> IOMMU IRTE with appropriate host physical APIC ID. Also, when
> vcpu_blocking/unblocking, SVM needs to update the is-running bit in
> the IOMMU IRTE. Both are achieved via calling amd_iommu_update_ga().
> 
> However, if GA mode is not enabled for the pass-through device,
> IOMMU driver will simply just return when calling amd_iommu_update_ga.
> 
> Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
> ---
> diff --git a/include/linux/amd-iommu.h b/include/linux/amd-iommu.h
> @@ -34,6 +34,7 @@ struct amd_ir_data {
>  	struct msi_msg msi_entry;
>  	void *entry;    /* Pointer to union irte or struct irte_ga */
>  	void *ref;      /* Pointer to the actual irte */
> +	struct list_head node;	/* Used by SVM for per-vcpu ir_list */

Putting a list_head here requires all users of amd-iommu to cooperate,
which is dangerous, but it allows simpler SVM code.  The alternative is
to force wrappers in SVM, which would also allow IOMMU to keep struct
amd_ir_data as incomplete in public headers.

 struct struct amd_ir_data_wrapper {
 	struct list_head node;
 	struct amd_ir_data *ir_data;
 }

(The rant continues below.)

> +static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
> +			      uint32_t guest_irq, bool set)
> +{
> +	struct kvm_kernel_irq_routing_entry *e;
> +	struct kvm_irq_routing_table *irq_rt;
[...]
> +	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
> +		struct kvm_lapic_irq irq;
> +		struct vcpu_data vcpu_info;
[...]
> +		kvm_set_msi_irq(e, &irq);
> +		if (kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
> +			svm = to_svm(vcpu);
> +			vcpu_info.pi_desc_addr = page_to_phys(svm->avic_backing_page);
> +			vcpu_info.vector = irq.vector;
[...]
> +			struct amd_iommu_pi_data pi;
> +
> +			/* Try to enable guest_mode in IRTE */
> +			pi.ga_tag = AVIC_GATAG(kvm->arch.avic_vm_id,
> +						     vcpu->vcpu_id);
> +			pi.is_guest_mode = true;
> +			pi.vcpu_data = &vcpu_info;
> +			ret = irq_set_vcpu_affinity(host_irq, &pi);
> +			if (!ret && pi.is_guest_mode)
> +				svm_ir_list_add(svm, pi.ir_data);

I missed a bug here the last time:

If ir_data is already inside some VCPU list and the VCPU changes, then
we don't remove ir_data from the previous list and just add it to a new
one.  This was not as bad when we only had wrappers (only resulted in
duplication), but now we break the removed list ...

The problem with wrappers is that we don't know what list we should
remove the "struct amd_ir_data" from;  we would need to add another
tracking structure or go through all VCPUs.

Having "struct list_head" in "struct amd_ir_data" would allow us to know
the current list and remove it from there:
One "struct amd_ir_data" should never be used by more than one caller of
amd_iommu_update_ga(), because they would have to be cooperating anyway,
which would mean a single mediator, so we can add a "struct list_head"
into "struct amd_ir_data".

  Minor design note:
  To make the usage of "struct amd_ir_data" safer, we could pass "struct
  list_head" into irq_set_vcpu_affinity(), instead of returning "struct
  amd_ir_data *".

  irq_set_vcpu_affinity() would add "struct amd_ir_data" to the list only
  if ir_data was not already in some list and report whether the list
  was modified.

I think that adding "struct list_head" into "struct amd_ir_data" is
nicer than having wrappers.

Joerg, Paolo, what do you think?

Thanks.

> diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
> @@ -4366,6 +4399,177 @@ static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec)
> +static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_ir_data *ir)
> +{
> +	spin_lock_irqsave(&svm->ir_list_lock, flags);
> +	list_for_each_entry(cur, &svm->ir_list, node) {
> +		if (cur != ir)
> +			continue;
> +		found = true;
> +		break;
> +	}

If we're using ir->node, then this can ask !list_empty(ir->node),
because we should never add it to a new list otherwise.

> +	spin_unlock_irqrestore(&svm->ir_list_lock, flags);
> +
> +	if (found)
> +		return 0;
> +
> +	spin_lock_irqsave(&svm->ir_list_lock, flags);
> +	list_add(&ir->node, &svm->ir_list);
> +	spin_unlock_irqrestore(&svm->ir_list_lock, flags);
> +	return 0;
> +}

Re: [PART2 PATCH v6 12/12] svm: Implements update_pi_irte hook to setup posted interrupt

[Radim Krčmář]

2016-08-19 16:49+0200, Radim Krčmář:
>   Minor design note:
>   To make the usage of "struct amd_ir_data" safer, we could pass "struct
>   list_head" into irq_set_vcpu_affinity(), instead of returning "struct
>   amd_ir_data *".
> 
>   irq_set_vcpu_affinity() would add "struct amd_ir_data" to the list only
>   if ir_data was not already in some list and report whether the list
>   was modified.

No, this would not simplify the search.  An important point of list_head
in amd_ir_data was to let us know the current list.

Re: [PART2 PATCH v6 07/12] iommu/amd: Introduce amd_iommu_update_ga()

[kbuild test robot]

[-- Attachment #1: Type: text/plain, Size: 1658 bytes --]

Hi Suravee,

[auto build test ERROR on v4.7-rc7]
[also build test ERROR on next-20160819]
[cannot apply to iommu/next kvm/linux-next linux/master v4.8-rc2 v4.8-rc1]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]
[Suggest to use git(>=2.9.0) format-patch --base=<commit> (or --base=auto for convenience) to record what (public, well-known) commit your patch series was built on]
[Check https://git-scm.com/docs/git-format-patch for more information]

url:    https://github.com/0day-ci/linux/commits/Suravee-Suthikulpanit/iommu-AMD-Introduce-IOMMU-AVIC-support/20160819-111539
config: x86_64-randconfig-x018-201633 (attached as .config)
compiler: gcc-6 (Debian 6.1.1-9) 6.1.1 20160705
reproduce:
        # save the attached .config to linux build tree
        make ARCH=x86_64 

All errors (new ones prefixed by >>):

   In file included from drivers/iommu/amd_iommu.c:33:0:
>> include/linux/amd-iommu.h:208:1: error: expected identifier or '(' before '{' token
    {
    ^
   include/linux/amd-iommu.h:207:1: warning: 'amd_iommu_update_ga' declared 'static' but never defined [-Wunused-function]
    amd_iommu_update_ga(u32 cpu, u64 base, bool is_run, struct amd_ir_data *ir_data);
    ^~~~~~~~~~~~~~~~~~~

vim +208 include/linux/amd-iommu.h

   202	{
   203		return 0;
   204	}
   205	
   206	static inline int
   207	amd_iommu_update_ga(u32 cpu, u64 base, bool is_run, struct amd_ir_data *ir_data);
 > 208	{
   209		return 0;
   210	}
   211	

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation

[-- Attachment #2: .config.gz --]
[-- Type: application/octet-stream, Size: 26798 bytes --]

Re: [PART2 PATCH v6 12/12] svm: Implements update_pi_irte hook to setup posted interrupt

[Suravee Suthikulpanit]

Hi Radim

On 08/19/2016 09:49 PM, Radim Krčmář wrote:
> 2016-08-18 14:42-0500, Suravee Suthikulpanit:
>> This patch implements update_pi_irte function hook to allow SVM
>> communicate to IOMMU driver regarding how to set up IRTE for handling
>> posted interrupt.
>>
>> In case AVIC is enabled, during vcpu_load/unload, SVM needs to update
>> IOMMU IRTE with appropriate host physical APIC ID. Also, when
>> vcpu_blocking/unblocking, SVM needs to update the is-running bit in
>> the IOMMU IRTE. Both are achieved via calling amd_iommu_update_ga().
>>
>> However, if GA mode is not enabled for the pass-through device,
>> IOMMU driver will simply just return when calling amd_iommu_update_ga.
>>
>> Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
>> ---
>> diff --git a/include/linux/amd-iommu.h b/include/linux/amd-iommu.h
>> @@ -34,6 +34,7 @@ struct amd_ir_data {
>>   	struct msi_msg msi_entry;
>>   	void *entry;    /* Pointer to union irte or struct irte_ga */
>>   	void *ref;      /* Pointer to the actual irte */
>> +	struct list_head node;	/* Used by SVM for per-vcpu ir_list */
>
> Putting a list_head here requires all users of amd-iommu to cooperate,
> which is dangerous, but it allows simpler SVM code.  The alternative is
> to force wrappers in SVM, which would also allow IOMMU to keep struct
> amd_ir_data as incomplete in public headers.
>
>   struct struct amd_ir_data_wrapper {
>   	struct list_head node;
>   	struct amd_ir_data *ir_data;
>   }
>
> (The rant continues below.)

The struct amd_ir_data is only used by AMD IOMMU driver (and now SVM 
driver). I don't think it would be that bad since they both would have 
to already coordinate.  Could you please clarify the point you mention 
that it could be dangerous?  Are you thinking of the case where the 
struct amd_ir_data could be free (by AMD IOMMU driver), while still in 
the vcpu list (in SVM)?

>> +static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
>> +			      uint32_t guest_irq, bool set)
>> +{
>> +	struct kvm_kernel_irq_routing_entry *e;
>> +	struct kvm_irq_routing_table *irq_rt;
> [...]
>> +	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
>> +		struct kvm_lapic_irq irq;
>> +		struct vcpu_data vcpu_info;
> [...]
>> +		kvm_set_msi_irq(e, &irq);
>> +		if (kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
>> +			svm = to_svm(vcpu);
>> +			vcpu_info.pi_desc_addr = page_to_phys(svm->avic_backing_page);
>> +			vcpu_info.vector = irq.vector;
> [...]
>> +			struct amd_iommu_pi_data pi;
>> +
>> +			/* Try to enable guest_mode in IRTE */
>> +			pi.ga_tag = AVIC_GATAG(kvm->arch.avic_vm_id,
>> +						     vcpu->vcpu_id);
>> +			pi.is_guest_mode = true;
>> +			pi.vcpu_data = &vcpu_info;
>> +			ret = irq_set_vcpu_affinity(host_irq, &pi);
>> +			if (!ret && pi.is_guest_mode)
>> +				svm_ir_list_add(svm, pi.ir_data);
>
> I missed a bug here the last time:
>
> If ir_data is already inside some VCPU list and the VCPU changes, then
> we don't remove ir_data from the previous list and just add it to a new
> one.  This was not as bad when we only had wrappers (only resulted in
> duplication), but now we break the removed list ...

Good point here...

> The problem with wrappers is that we don't know what list we should
> remove the "struct amd_ir_data" from;  we would need to add another
> tracking structure or go through all VCPUs.
>
> Having "struct list_head" in "struct amd_ir_data" would allow us to know
> the current list and remove it from there:
> One "struct amd_ir_data" should never be used by more than one caller of
> amd_iommu_update_ga(), because they would have to be cooperating anyway,
> which would mean a single mediator, so we can add a "struct list_head"
> into "struct amd_ir_data".
>
>    Minor design note:
>    To make the usage of "struct amd_ir_data" safer, we could pass "struct
>    list_head" into irq_set_vcpu_affinity(), instead of returning "struct
>    amd_ir_data *".
>
>    irq_set_vcpu_affinity() would add "struct amd_ir_data" to the list only
>    if ir_data was not already in some list and report whether the list
>    was modified.
>
> I think that adding "struct list_head" into "struct amd_ir_data" is
> nicer than having wrappers.
>
> Joerg, Paolo, what do you think?
>

I think modifying irq_set_vcpu_affinity() to also pass struct list_head 
seems a bit redundant since it is currently design to allow passing in 
void *, which leaves the other option where we might just need to pass 
in a wrapper (e.g. going back to the previous design where we pass in 
struct amd_iommu_pi_data) and also add a pointer to the ir_list in the 
wrapper as well. Then, IOMMU is responsible for adding/deleting ir_data 
to/from this list instead of SVM. This should be fine since we only need 
to coordinate b/w SVM and AMD-IOMMU.

Thanks,
Suravee

Re: [PART2 PATCH v6 12/12] svm: Implements update_pi_irte hook to setup posted interrupt

[Suravee Suthikulpanit]

Hi Radim,

On 08/22/2016 04:19 PM, Suravee Suthikulpanit wrote:
>> he problem with wrappers is that we don't know what list we should
>> remove the "struct amd_ir_data" from;  we would need to add another
>> tracking structure or go through all VCPUs.
>>
>> Having "struct list_head" in "struct amd_ir_data" would allow us to know
>> the current list and remove it from there:
>> One "struct amd_ir_data" should never be used by more than one caller of
>> amd_iommu_update_ga(), because they would have to be cooperating anyway,
>> which would mean a single mediator, so we can add a "struct list_head"
>> into "struct amd_ir_data".
>>
>>    Minor design note:
>>    To make the usage of "struct amd_ir_data" safer, we could pass "struct
>>    list_head" into irq_set_vcpu_affinity(), instead of returning "struct
>>    amd_ir_data *".
>>
>>    irq_set_vcpu_affinity() would add "struct amd_ir_data" to the list
>> only
>>    if ir_data was not already in some list and report whether the list
>>    was modified.
>>
>> I think that adding "struct list_head" into "struct amd_ir_data" is
>> nicer than having wrappers.
>>
>> Joerg, Paolo, what do you think?
>>
>
> I think modifying irq_set_vcpu_affinity() to also pass struct list_head
> seems a bit redundant since it is currently design to allow passing in
> void *, which leaves the other option where we might just need to pass
> in a wrapper (e.g. going back to the previous design where we pass in
> struct amd_iommu_pi_data) and also add a pointer to the ir_list in the
> wrapper as well. Then, IOMMU is responsible for adding/deleting ir_data
> to/from this list instead of SVM. This should be fine since we only need
> to coordinate b/w SVM and AMD-IOMMU.
>
> Thanks,
> Suravee

Actually, thinking about this again, going back to keeping the per-vcpu 
list of struct amd_iommu_pi_data is probably the simplest here.

* We avoid having to expose the amd_ir_data to SVM.
* We can match using amd_ir_data * when traversing the list.
* We can easily add the code to manage the list in the SVM. We can make 
sure that the struct amd_iommu_pi_data is not already mapped before 
adding it to a new per-vcpu list. If it is currently mapped, we can 
simply unmapped it. Doing this from IOMMU would be more complicate and 
require lots of parameter passing.

Thanks,
Suravee

Re: [PART2 PATCH v6 12/12] svm: Implements update_pi_irte hook to setup posted interrupt

[Radim Krčmář]

2016-08-22 17:09+0700, Suravee Suthikulpanit:
> On 08/22/2016 04:19 PM, Suravee Suthikulpanit wrote:
>> > he problem with wrappers is that we don't know what list we should
>> > remove the "struct amd_ir_data" from;  we would need to add another
>> > tracking structure or go through all VCPUs.
>> > 
>> > Having "struct list_head" in "struct amd_ir_data" would allow us to know
>> > the current list and remove it from there:
>> > One "struct amd_ir_data" should never be used by more than one caller of
>> > amd_iommu_update_ga(), because they would have to be cooperating anyway,
>> > which would mean a single mediator, so we can add a "struct list_head"
>> > into "struct amd_ir_data".
>> > 
>> >    Minor design note:
>> >    To make the usage of "struct amd_ir_data" safer, we could pass "struct
>> >    list_head" into irq_set_vcpu_affinity(), instead of returning "struct
>> >    amd_ir_data *".
>> > 
>> >    irq_set_vcpu_affinity() would add "struct amd_ir_data" to the list
>> > only
>> >    if ir_data was not already in some list and report whether the list
>> >    was modified.
>> > 
>> > I think that adding "struct list_head" into "struct amd_ir_data" is
>> > nicer than having wrappers.
>> > 
>> > Joerg, Paolo, what do you think?
>> > 
>> 
>> I think modifying irq_set_vcpu_affinity() to also pass struct list_head
>> seems a bit redundant since it is currently design to allow passing in
>> void *, which leaves the other option where we might just need to pass
>> in a wrapper (e.g. going back to the previous design where we pass in
>> struct amd_iommu_pi_data) and also add a pointer to the ir_list in the
>> wrapper as well. Then, IOMMU is responsible for adding/deleting ir_data
>> to/from this list instead of SVM. This should be fine since we only need
>> to coordinate b/w SVM and AMD-IOMMU.
> 
> Actually, thinking about this again, going back to keeping the per-vcpu list
> of struct amd_iommu_pi_data is probably the simplest here.
> 
> * We avoid having to expose the amd_ir_data to SVM.
> * We can match using amd_ir_data * when traversing the list.
> * We can easily add the code to manage the list in the SVM. We can make sure
> that the struct amd_iommu_pi_data is not already mapped before adding it to
> a new per-vcpu list. If it is currently mapped, we can simply unmapped it.

Sounds good.
A new SVM-specific wrapper for amd_ir_data instead of reusing
amd_iommu_pi_data would be nicer, IMO -- we could change it without
touching the IOMMU interface and also allocate in svm_pi_list_add.

Updating lists would become O(N^2), but updates should not occur often
(and N small) so I think it's still worth the saving on every
sched_in/out.

> Doing this from IOMMU would be more complicate and require lots of parameter
> passing.

Yeah, doing more than returning amd_ir_data from IOMMU doesn't make
sense and not adding list_head for SVm to amd_ir_data is more
acceptable.

Thanks.

Re: [PART2 PATCH v6 12/12] svm: Implements update_pi_irte hook to setup posted interrupt

[kbuild test robot]

[-- Attachment #1: Type: text/plain, Size: 2390 bytes --]

Hi Suravee,

[auto build test WARNING on v4.7-rc7]
[cannot apply to iommu/next kvm/linux-next linux/master v4.8-rc2 v4.8-rc1 next-20160823]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]
[Suggest to use git(>=2.9.0) format-patch --base=<commit> (or --base=auto for convenience) to record what (public, well-known) commit your patch series was built on]
[Check https://git-scm.com/docs/git-format-patch for more information]

url:    https://github.com/0day-ci/linux/commits/Suravee-Suthikulpanit/iommu-AMD-Introduce-IOMMU-AVIC-support/20160819-111539
config: i386-randconfig-s0-201633 (attached as .config)
compiler: gcc-6 (Debian 6.1.1-9) 6.1.1 20160705
reproduce:
        # save the attached .config to linux build tree
        make ARCH=i386 

All warnings (new ones prefixed by >>):

   In file included from arch/x86/kvm/svm.c:37:0:
   include/linux/amd-iommu.h:222:1: error: expected identifier or '(' before '{' token
    {
    ^
>> include/linux/amd-iommu.h:221:1: warning: 'amd_iommu_update_ga' used but never defined
    amd_iommu_update_ga(u32 cpu, u64 base, bool is_run, struct amd_ir_data *ir_data);
    ^~~~~~~~~~~~~~~~~~~

vim +/amd_iommu_update_ga +221 include/linux/amd-iommu.h

bda901f5 Suravee Suthikulpanit 2016-08-18  215  amd_iommu_register_ga_log_notifier(int (*notifier)(u32))
bda901f5 Suravee Suthikulpanit 2016-08-18  216  {
bda901f5 Suravee Suthikulpanit 2016-08-18  217  	return 0;
bda901f5 Suravee Suthikulpanit 2016-08-18  218  }
bda901f5 Suravee Suthikulpanit 2016-08-18  219  
d49c1352 Suravee Suthikulpanit 2016-08-18  220  static inline int
d49c1352 Suravee Suthikulpanit 2016-08-18 @221  amd_iommu_update_ga(u32 cpu, u64 base, bool is_run, struct amd_ir_data *ir_data);
d49c1352 Suravee Suthikulpanit 2016-08-18 @222  {
d49c1352 Suravee Suthikulpanit 2016-08-18  223  	return 0;
d49c1352 Suravee Suthikulpanit 2016-08-18  224  }
d49c1352 Suravee Suthikulpanit 2016-08-18  225  

:::::: The code at line 221 was first introduced by commit
:::::: d49c1352bd32284fa9619171c8a505973cf79c54 iommu/amd: Introduce amd_iommu_update_ga()

:::::: TO: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
:::::: CC: 0day robot <fengguang.wu@intel.com>

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation

[-- Attachment #2: .config.gz --]
[-- Type: application/octet-stream, Size: 31358 bytes --]