[patch 1/4] dmar: support for parsing Remapping Hardware Static Affinity structure

[patch 1/4] dmar: support for parsing Remapping Hardware Static Affinity structure

[Suresh Siddha]

[-- Attachment #1: parse_rhsa_struct.patch --]
[-- Type: text/plain, Size: 3733 bytes --]

Add support for parsing Remapping Hardware Static Affinity (RHSA) structure.
This enables identifying the association between remapping hardware units and
the corresponding proximity domain. This enables to allocate transalation
structures closer to the remapping hardware unit.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
---
 drivers/pci/dmar.c          |   32 ++++++++++++++++++++++++++++++++
 include/acpi/actbl1.h       |   12 +++++++++++-
 include/linux/intel-iommu.h |    1 +
 3 files changed, 44 insertions(+), 1 deletion(-)

Index: tip/drivers/pci/dmar.c
===================================================================
--- tip.orig/drivers/pci/dmar.c
+++ tip/drivers/pci/dmar.c
@@ -348,12 +348,33 @@ found:
 }
 #endif
 
+static int __init
+dmar_parse_one_rhsa(struct acpi_dmar_header *header)
+{
+	struct acpi_dmar_rhsa *rhsa;
+	struct dmar_drhd_unit *drhd;
+
+	rhsa = (struct acpi_dmar_rhsa *)header;
+	for_each_drhd_unit(drhd)
+		if (drhd->reg_base_addr == rhsa->base_address) {
+			int node = acpi_map_pxm_to_node(rhsa->pxm);
+
+			if (!node_online(node))
+				node = -1;
+			drhd->iommu->node = node;
+			return 0;
+		}
+
+	return -ENODEV;
+}
+
 static void __init
 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
 {
 	struct acpi_dmar_hardware_unit *drhd;
 	struct acpi_dmar_reserved_memory *rmrr;
 	struct acpi_dmar_atsr *atsr;
+	struct acpi_dmar_rhsa *rhsa;
 
 	switch (header->type) {
 	case ACPI_DMAR_TYPE_HARDWARE_UNIT:
@@ -375,6 +396,12 @@ dmar_table_print_dmar_entry(struct acpi_
 		atsr = container_of(header, struct acpi_dmar_atsr, header);
 		printk(KERN_INFO PREFIX "ATSR flags: %#x\n", atsr->flags);
 		break;
+	case ACPI_DMAR_TYPE_RHSA:
+		rhsa = container_of(header, struct acpi_dmar_rhsa, header);
+		printk(KERN_INFO PREFIX
+		       "RHSA base: %#016Lx pxm: %#x\n",
+		       rhsa->base_address, rhsa->pxm);
+		break;
 	}
 }
 
@@ -459,6 +486,9 @@ parse_dmar_table(void)
 			ret = dmar_parse_one_atsr(entry_header);
 #endif
 			break;
+		case ACPI_DMAR_TYPE_RHSA:
+			ret = dmar_parse_one_rhsa(entry_header);
+			break;
 		default:
 			printk(KERN_WARNING PREFIX
 				"Unknown DMAR structure type\n");
@@ -658,6 +688,8 @@ int alloc_iommu(struct dmar_drhd_unit *d
 	iommu->agaw = agaw;
 	iommu->msagaw = msagaw;
 
+	iommu->node = -1;
+
 	/* the registers might be more than one page */
 	map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
 		cap_max_fault_reg_offset(iommu->cap));
Index: tip/include/acpi/actbl1.h
===================================================================
--- tip.orig/include/acpi/actbl1.h
+++ tip/include/acpi/actbl1.h
@@ -328,7 +328,8 @@ enum acpi_dmar_type {
 	ACPI_DMAR_TYPE_HARDWARE_UNIT = 0,
 	ACPI_DMAR_TYPE_RESERVED_MEMORY = 1,
 	ACPI_DMAR_TYPE_ATSR = 2,
-	ACPI_DMAR_TYPE_RESERVED = 3	/* 3 and greater are reserved */
+	ACPI_DMAR_TYPE_RHSA = 3,
+	ACPI_DMAR_TYPE_RESERVED = 4	/* 4 and greater are reserved */
 };
 
 struct acpi_dmar_device_scope {
@@ -401,6 +402,15 @@ struct acpi_dmar_atsr {
 
 #define ACPI_DMAR_ALL_PORTS	    (1)
 
+/* 3: Remapping Hardware Static Affinity Reporting Structure */
+
+struct acpi_dmar_rhsa {
+	struct acpi_dmar_header header;
+	u32 reserved;
+	u64 base_address;
+	u32 pxm;
+};
+
 /*******************************************************************************
  *
  * ECDT - Embedded Controller Boot Resources Table
Index: tip/include/linux/intel-iommu.h
===================================================================
--- tip.orig/include/linux/intel-iommu.h
+++ tip/include/linux/intel-iommu.h
@@ -332,6 +332,7 @@ struct intel_iommu {
 #ifdef CONFIG_INTR_REMAP
 	struct ir_table *ir_table;	/* Interrupt remapping info */
 #endif
+	int		node;
 };
 
 static inline void __iommu_flush_cache(

[patch 2/4] dmar: Allocate queued invalidation structure using numa locality info

[Suresh Siddha]

[-- Attachment #1: qi_use_affinity_info.patch --]
[-- Type: text/plain, Size: 1078 bytes --]

Allocate queued invalidation descriptor structures using numa locality info.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
---
 drivers/pci/dmar.c |    8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

Index: tip/drivers/pci/dmar.c
===================================================================
--- tip.orig/drivers/pci/dmar.c
+++ tip/drivers/pci/dmar.c
@@ -1028,6 +1028,7 @@ static void __dmar_enable_qi(struct inte
 int dmar_enable_qi(struct intel_iommu *iommu)
 {
 	struct q_inval *qi;
+	struct page *desc_page;
 
 	if (!ecap_qis(iommu->ecap))
 		return -ENOENT;
@@ -1044,13 +1045,16 @@ int dmar_enable_qi(struct intel_iommu *i
 
 	qi = iommu->qi;
 
-	qi->desc = (void *)(get_zeroed_page(GFP_ATOMIC));
-	if (!qi->desc) {
+
+	desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 0);
+	if (!desc_page) {
 		kfree(qi);
 		iommu->qi = 0;
 		return -ENOMEM;
 	}
 
+	qi->desc = page_address(desc_page);
+
 	qi->desc_status = kmalloc(QI_LENGTH * sizeof(int), GFP_ATOMIC);
 	if (!qi->desc_status) {
 		free_page((unsigned long) qi->desc);

[patch 3/4] intr_remap: Allocate intr-remapping table using numa locality info

[Suresh Siddha]

[-- Attachment #1: intr_remap_use_affinity_info.patch --]
[-- Type: text/plain, Size: 895 bytes --]

Allocate intr-remapping table using numa locality info. On platforms
having remapping hardware units span different nodes, this enables
optimized intr-remapping table entry access by remapping hardware.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
---
 drivers/pci/intr_remapping.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Index: tip/drivers/pci/intr_remapping.c
===================================================================
--- tip.orig/drivers/pci/intr_remapping.c
+++ tip/drivers/pci/intr_remapping.c
@@ -590,7 +590,8 @@ static int setup_intr_remapping(struct i
 	if (!iommu->ir_table)
 		return -ENOMEM;
 
-	pages = alloc_pages(GFP_ATOMIC | __GFP_ZERO, INTR_REMAP_PAGE_ORDER);
+	pages = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO,
+				 INTR_REMAP_PAGE_ORDER);
 
 	if (!pages) {
 		printk(KERN_ERR "failed to allocate pages of order %d\n",

[patch 4/4] iommu: Allocate dma-remapping structures using numa locality info

[Suresh Siddha]

[-- Attachment #1: iommu_use_affinity_info.patch --]
[-- Type: text/plain, Size: 3901 bytes --]

Allocate dma-remapping structures using numa locality info. On platforms
having remapping hardware units span different nodes, this enables
optimized dma-remapping transalation structures access by remapping hardware.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
---
 drivers/pci/intel-iommu.c |   27 +++++++++++++++++++--------
 1 file changed, 19 insertions(+), 8 deletions(-)

Index: tip/drivers/pci/intel-iommu.c
===================================================================
--- tip.orig/drivers/pci/intel-iommu.c
+++ tip/drivers/pci/intel-iommu.c
@@ -267,6 +267,7 @@ struct dmar_domain *si_domain;
 
 struct dmar_domain {
 	int	id;			/* domain id */
+	int	nid;			/* node id */
 	unsigned long iommu_bmp;	/* bitmap of iommus this domain uses*/
 
 	struct list_head devices; 	/* all devices' list */
@@ -390,15 +391,18 @@ static inline void *iommu_kmem_cache_all
 }
 
 
-static inline void *alloc_pgtable_page(void)
+static inline void *alloc_pgtable_page(int node)
 {
 	unsigned int flags;
-	void *vaddr;
+	struct page *page;
+	void *vaddr = NULL;
 
 	/* trying to avoid low memory issues */
 	flags = current->flags & PF_MEMALLOC;
 	current->flags |= PF_MEMALLOC;
-	vaddr = (void *)get_zeroed_page(GFP_ATOMIC);
+	page = alloc_pages_node(node, GFP_ATOMIC | __GFP_ZERO, 0);
+	if (page)
+		vaddr = page_address(page);
 	current->flags &= (~PF_MEMALLOC | flags);
 	return vaddr;
 }
@@ -579,7 +583,8 @@ static struct context_entry * device_to_
 	root = &iommu->root_entry[bus];
 	context = get_context_addr_from_root(root);
 	if (!context) {
-		context = (struct context_entry *)alloc_pgtable_page();
+		context = (struct context_entry *)
+				alloc_pgtable_page(iommu->node);
 		if (!context) {
 			spin_unlock_irqrestore(&iommu->lock, flags);
 			return NULL;
@@ -722,7 +727,7 @@ static struct dma_pte *pfn_to_dma_pte(st
 		if (!dma_pte_present(pte)) {
 			uint64_t pteval;
 
-			tmp_page = alloc_pgtable_page();
+			tmp_page = alloc_pgtable_page(domain->nid);
 
 			if (!tmp_page)
 				return NULL;
@@ -855,7 +860,7 @@ static int iommu_alloc_root_entry(struct
 	struct root_entry *root;
 	unsigned long flags;
 
-	root = (struct root_entry *)alloc_pgtable_page();
+	root = (struct root_entry *)alloc_pgtable_page(iommu->node);
 	if (!root)
 		return -ENOMEM;
 
@@ -1249,6 +1254,7 @@ static struct dmar_domain *alloc_domain(
 	if (!domain)
 		return NULL;
 
+	domain->nid = -1;
 	memset(&domain->iommu_bmp, 0, sizeof(unsigned long));
 	domain->flags = 0;
 
@@ -1409,9 +1415,10 @@ static int domain_init(struct dmar_domai
 		domain->iommu_snooping = 0;
 
 	domain->iommu_count = 1;
+	domain->nid = iommu->node;
 
 	/* always allocate the top pgd */
-	domain->pgd = (struct dma_pte *)alloc_pgtable_page();
+	domain->pgd = (struct dma_pte *)alloc_pgtable_page(domain->nid);
 	if (!domain->pgd)
 		return -ENOMEM;
 	__iommu_flush_cache(iommu, domain->pgd, PAGE_SIZE);
@@ -1566,6 +1573,8 @@ static int domain_context_mapping_one(st
 	spin_lock_irqsave(&domain->iommu_lock, flags);
 	if (!test_and_set_bit(iommu->seq_id, &domain->iommu_bmp)) {
 		domain->iommu_count++;
+		if (domain->iommu_count == 1)
+			domain->nid = iommu->node;
 		domain_update_iommu_cap(domain);
 	}
 	spin_unlock_irqrestore(&domain->iommu_lock, flags);
@@ -3397,6 +3406,7 @@ static struct dmar_domain *iommu_alloc_v
 		return NULL;
 
 	domain->id = vm_domid++;
+	domain->nid = -1;
 	memset(&domain->iommu_bmp, 0, sizeof(unsigned long));
 	domain->flags = DOMAIN_FLAG_VIRTUAL_MACHINE;
 
@@ -3423,9 +3433,10 @@ static int md_domain_init(struct dmar_do
 	domain->iommu_coherency = 0;
 	domain->iommu_snooping = 0;
 	domain->max_addr = 0;
+	domain->nid = -1;
 
 	/* always allocate the top pgd */
-	domain->pgd = (struct dma_pte *)alloc_pgtable_page();
+	domain->pgd = (struct dma_pte *)alloc_pgtable_page(domain->nid);
 	if (!domain->pgd)
 		return -ENOMEM;
 	domain_flush_cache(domain, domain->pgd, PAGE_SIZE);

Re: [patch 1/4] dmar: support for parsing Remapping Hardware Static Affinity structure

[David Woodhouse]

On Thu, 2009-10-01 at 16:06 -0700, Suresh Siddha wrote:
> Add support for parsing Remapping Hardware Static Affinity (RHSA) structure.
> This enables identifying the association between remapping hardware units and
> the corresponding proximity domain. This enables to allocate transalation
> structures closer to the remapping hardware unit.

That's not going to apply on top of commit 17b6097753e, is it?

-- 
David Woodhouse                            Open Source Technology Centre
David.Woodhouse@intel.com                              Intel Corporation

Re: [patch 1/4] dmar: support for parsing Remapping Hardware Static Affinity structure

[Suresh Siddha]

On Thu, 2009-10-01 at 22:39 -0700, David Woodhouse wrote:
> On Thu, 2009-10-01 at 16:06 -0700, Suresh Siddha wrote:
> > Add support for parsing Remapping Hardware Static Affinity (RHSA) structure.
> > This enables identifying the association between remapping hardware units and
> > the corresponding proximity domain. This enables to allocate transalation
> > structures closer to the remapping hardware unit.
> 
> That's not going to apply on top of commit 17b6097753e, is it?

Oops, didn't notice Roland's patch. Just posted the new set on top of
your tree.

thanks,
suresh

[patch 1/4] dmar: support for parsing Remapping Hardware Static Affinity structure

[Suresh Siddha]

[-- Attachment #1: parse_rhsa_struct.patch --]
[-- Type: text/plain, Size: 2162 bytes --]

Add support for parsing Remapping Hardware Static Affinity (RHSA) structure.
This enables identifying the association between remapping hardware units and
the corresponding proximity domain. This enables to allocate transalation
structures closer to the remapping hardware unit.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
---
 drivers/pci/dmar.c          |   24 +++++++++++++++++++++++-
 include/linux/intel-iommu.h |    1 +
 2 files changed, 24 insertions(+), 1 deletion(-)

Index: iommu-2.6/drivers/pci/dmar.c
===================================================================
--- iommu-2.6.orig/drivers/pci/dmar.c
+++ iommu-2.6/drivers/pci/dmar.c
@@ -348,6 +348,26 @@ found:
 }
 #endif
 
+static int __init
+dmar_parse_one_rhsa(struct acpi_dmar_header *header)
+{
+	struct acpi_dmar_rhsa *rhsa;
+	struct dmar_drhd_unit *drhd;
+
+	rhsa = (struct acpi_dmar_rhsa *)header;
+	for_each_drhd_unit(drhd)
+		if (drhd->reg_base_addr == rhsa->base_address) {
+			int node = acpi_map_pxm_to_node(rhsa->proximity_domain);
+
+			if (!node_online(node))
+				node = -1;
+			drhd->iommu->node = node;
+		return 0;
+	}
+
+	return -ENODEV;
+}
+
 static void __init
 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
 {
@@ -467,7 +487,7 @@ parse_dmar_table(void)
 #endif
 			break;
 		case ACPI_DMAR_HARDWARE_AFFINITY:
-			/* We don't do anything with RHSA (yet?) */
+			ret = dmar_parse_one_rhsa(entry_header);
 			break;
 		default:
 			printk(KERN_WARNING PREFIX
@@ -677,6 +697,8 @@ int alloc_iommu(struct dmar_drhd_unit *d
 	iommu->agaw = agaw;
 	iommu->msagaw = msagaw;
 
+	iommu->node = -1;
+
 	/* the registers might be more than one page */
 	map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
 		cap_max_fault_reg_offset(iommu->cap));
Index: iommu-2.6/include/linux/intel-iommu.h
===================================================================
--- iommu-2.6.orig/include/linux/intel-iommu.h
+++ iommu-2.6/include/linux/intel-iommu.h
@@ -332,6 +332,7 @@ struct intel_iommu {
 #ifdef CONFIG_INTR_REMAP
 	struct ir_table *ir_table;	/* Interrupt remapping info */
 #endif
+	int		node;
 };
 
 static inline void __iommu_flush_cache(