iommu/amd: Allocate page table using numa locality info

Introduce 'struct protection_domain->nid' variable. It will contain
IOMMU NUMA node ID. And allocate page table pages using IOMMU numa
locality info. This optimizes page table walk by IOMMU.
Signed-off-by: Vasant Hegde <vasant.hegde@amd.com>
Link: https://lore.kernel.org/r/20230321092348.6127-2-vasant.hegde@amd.comSigned-off-by: Joerg Roedel <jroedel@suse.de>

drivers/iommu/amd/amd_iommu.h

@@ -122,6 +122,14 @@ static inline int get_pci_sbdf_id(struct pci_dev *pdev)
 	return PCI_SEG_DEVID_TO_SBDF(seg, devid);
 }
 
+static inline void *alloc_pgtable_page(int nid, gfp_t gfp)
+{
+	struct page *page;
+
+	page = alloc_pages_node(nid, gfp | __GFP_ZERO, 0);
+	return page ? page_address(page) : NULL;
+}
+
 extern bool translation_pre_enabled(struct amd_iommu *iommu);
 extern bool amd_iommu_is_attach_deferred(struct device *dev);
 extern int __init add_special_device(u8 type, u8 id, u32 *devid,

drivers/iommu/amd/amd_iommu_types.h

@@ -549,6 +549,7 @@ struct protection_domain {
 	spinlock_t lock;	/* mostly used to lock the page table*/
 	u16 id;			/* the domain id written to the device table */
 	int glx;		/* Number of levels for GCR3 table */
+	int nid;		/* Node ID */
 	u64 *gcr3_tbl;		/* Guest CR3 table */
 	unsigned long flags;	/* flags to find out type of domain */
 	unsigned dev_cnt;	/* devices assigned to this domain */

drivers/iommu/amd/io_pgtable.c

@@ -156,7 +156,7 @@ static bool increase_address_space(struct protection_domain *domain,
 	bool ret = true;
 	u64 *pte;
 
-	pte = (void *)get_zeroed_page(gfp);
+	pte = alloc_pgtable_page(domain->nid, gfp);
 	if (!pte)
 		return false;
 
@@ -250,7 +250,7 @@ static u64 *alloc_pte(struct protection_domain *domain,
 
 		if (!IOMMU_PTE_PRESENT(__pte) ||
 		    pte_level == PAGE_MODE_NONE) {
-			page = (u64 *)get_zeroed_page(gfp);
+			page = alloc_pgtable_page(domain->nid, gfp);
 
 			if (!page)
 				return NULL;

drivers/iommu/amd/io_pgtable_v2.c

@@ -46,11 +46,6 @@ static inline bool is_large_pte(u64 pte)
 	return (pte & IOMMU_PAGE_PSE);
 }
 
-static inline void *alloc_pgtable_page(void)
-{
-	return (void *)get_zeroed_page(GFP_KERNEL);
-}
-
 static inline u64 set_pgtable_attr(u64 *page)
 {
 	u64 prot;
@@ -138,8 +133,8 @@ static void free_pgtable(u64 *pt, int level)
 }
 
 /* Allocate page table */
-static u64 *v2_alloc_pte(u64 *pgd, unsigned long iova,
-			 unsigned long pg_size, bool *updated)
+static u64 *v2_alloc_pte(int nid, u64 *pgd, unsigned long iova,
+			 unsigned long pg_size, gfp_t gfp, bool *updated)
 {
 	u64 *pte, *page;
 	int level, end_level;
@@ -162,7 +157,7 @@ static u64 *v2_alloc_pte(u64 *pgd, unsigned long iova,
 		}
 
 		if (!IOMMU_PTE_PRESENT(__pte)) {
-			page = alloc_pgtable_page();
+			page = alloc_pgtable_page(nid, gfp);
 			if (!page)
 				return NULL;
 
@@ -262,7 +257,8 @@ static int iommu_v2_map_pages(struct io_pgtable_ops *ops, unsigned long iova,
 
 	while (mapped_size < size) {
 		map_size = get_alloc_page_size(pgsize);
-		pte = v2_alloc_pte(pdom->iop.pgd, iova, map_size, &updated);
+		pte = v2_alloc_pte(pdom->nid, pdom->iop.pgd,
+				   iova, map_size, gfp, &updated);
 		if (!pte) {
 			ret = -EINVAL;
 			goto out;
@@ -384,7 +380,7 @@ static struct io_pgtable *v2_alloc_pgtable(struct io_pgtable_cfg *cfg, void *coo
 	struct protection_domain *pdom = (struct protection_domain *)cookie;
 	int ret;
 
-	pgtable->pgd = alloc_pgtable_page();
+	pgtable->pgd = alloc_pgtable_page(pdom->nid, GFP_ATOMIC);
 	if (!pgtable->pgd)
 		return NULL;
 

drivers/iommu/amd/iommu.c

@@ -1662,6 +1662,10 @@ static void do_attach(struct iommu_dev_data *dev_data,
 	dev_data->domain = domain;
 	list_add(&dev_data->list, &domain->dev_list);
 
+	/* Update NUMA Node ID */
+	if (domain->nid == NUMA_NO_NODE)
+		domain->nid = dev_to_node(dev_data->dev);
+
 	/* Do reference counting */
 	domain->dev_iommu[iommu->index] += 1;
 	domain->dev_cnt                 += 1;
@@ -2097,6 +2101,8 @@ static struct protection_domain *protection_domain_alloc(unsigned int type)
 	if (type == IOMMU_DOMAIN_IDENTITY)
 		return domain;
 
+	domain->nid = NUMA_NO_NODE;
+
 	pgtbl_ops = alloc_io_pgtable_ops(pgtable, &domain->iop.pgtbl_cfg, domain);
 	if (!pgtbl_ops) {
 		domain_id_free(domain->id);