iommu/s390: support additional table regions

index e1c76e0f9c2b..cad032d4c9a6 100644
@@ -31,6 +31,7 @@ struct s390_domain {
 	unsigned long		*dma_table;
 	spinlock_t		list_lock;
 	struct rcu_head		rcu;
+	u8			origin_type;
 };
 
 static struct iommu_domain blocking_domain;

@@ -345,6 +346,7 @@ static struct iommu_domain *s390_domain_alloc_paging(struct device *dev)
 	s390_domain->domain.geometry.force_aperture = true;
 	s390_domain->domain.geometry.aperture_start = 0;
 	s390_domain->domain.geometry.aperture_end = ZPCI_TABLE_SIZE_RT - 1;
+	s390_domain->origin_type = ZPCI_TABLE_TYPE_RTX;
 
 	spin_lock_init(&s390_domain->list_lock);
 	INIT_LIST_HEAD_RCU(&s390_domain->devices);

@@ -381,6 +383,21 @@ static void zdev_s390_domain_update(struct zpci_dev *zdev,
 	spin_unlock_irqrestore(&zdev->dom_lock, flags);
 }
 
+static u64 get_iota_region_flag(struct s390_domain *domain)
+{
+	switch (domain->origin_type) {
+	case ZPCI_TABLE_TYPE_RTX:
+		return ZPCI_IOTA_RTTO_FLAG;
+	case ZPCI_TABLE_TYPE_RSX:
+		return ZPCI_IOTA_RSTO_FLAG;
+	case ZPCI_TABLE_TYPE_RFX:
+		return ZPCI_IOTA_RFTO_FLAG;
+	default:
+		WARN_ONCE(1, "Invalid IOMMU table (%x)\n", domain->origin_type);
+		return 0;
+	}
+}
+
 static int s390_iommu_domain_reg_ioat(struct zpci_dev *zdev,
 				      struct iommu_domain *domain, u8 *status)
 {

@@ -399,7 +416,7 @@ static int s390_iommu_domain_reg_ioat(struct zpci_dev *zdev,
 	default:
 		s390_domain = to_s390_domain(domain);
 		iota = virt_to_phys(s390_domain->dma_table) |
-		       ZPCI_IOTA_RTTO_FLAG;
+		       get_iota_region_flag(s390_domain);
 		rc = zpci_register_ioat(zdev, 0, zdev->start_dma,
 					zdev->end_dma, iota, status);
 	}

index cad032d4c9a6..f2cda0ce0fe9 100644
@@ -121,6 +121,22 @@ static inline int pt_entry_isvalid(unsigned long entry)
 	return (entry & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID;
 }
 
+static inline unsigned long *get_rf_rso(unsigned long entry)
+{
+	if ((entry & ZPCI_TABLE_TYPE_MASK) == ZPCI_TABLE_TYPE_RFX)
+		return phys_to_virt(entry & ZPCI_RTE_ADDR_MASK);
+	else
+		return NULL;
+}
+
+static inline unsigned long *get_rs_rto(unsigned long entry)
+{
+	if ((entry & ZPCI_TABLE_TYPE_MASK) == ZPCI_TABLE_TYPE_RSX)
+		return phys_to_virt(entry & ZPCI_RTE_ADDR_MASK);
+	else
+		return NULL;
+}
+
 static inline unsigned long *get_rt_sto(unsigned long entry)
 {
 	if ((entry & ZPCI_TABLE_TYPE_MASK) == ZPCI_TABLE_TYPE_RTX)

@@ -192,18 +208,59 @@ static void dma_free_seg_table(unsigned long entry)
 	dma_free_cpu_table(sto);
 }
 
-static void dma_cleanup_tables(unsigned long *table)
+static void dma_free_rt_table(unsigned long entry)
 {
+	unsigned long *rto = get_rs_rto(entry);
 	int rtx;
 
-	if (!table)
+	for (rtx = 0; rtx < ZPCI_TABLE_ENTRIES; rtx++)
+		if (reg_entry_isvalid(rto[rtx]))
+			dma_free_seg_table(rto[rtx]);
+
+	dma_free_cpu_table(rto);
+}
+
+static void dma_free_rs_table(unsigned long entry)
+{
+	unsigned long *rso = get_rf_rso(entry);
+	int rsx;
+
+	for (rsx = 0; rsx < ZPCI_TABLE_ENTRIES; rsx++)
+		if (reg_entry_isvalid(rso[rsx]))
+			dma_free_rt_table(rso[rsx]);
+
+	dma_free_cpu_table(rso);
+}
+
+static void dma_cleanup_tables(struct s390_domain *domain)
+{
+	int rtx, rsx, rfx;
+
+	if (!domain->dma_table)
 		return;
 
-	for (rtx = 0; rtx < ZPCI_TABLE_ENTRIES; rtx++)
-		if (reg_entry_isvalid(table[rtx]))
-			dma_free_seg_table(table[rtx]);
+	switch (domain->origin_type) {
+	case ZPCI_TABLE_TYPE_RFX:
+		for (rfx = 0; rfx < ZPCI_TABLE_ENTRIES; rfx++)
+			if (reg_entry_isvalid(domain->dma_table[rfx]))
+				dma_free_rs_table(domain->dma_table[rfx]);
+		break;
+	case ZPCI_TABLE_TYPE_RSX:
+		for (rsx = 0; rsx < ZPCI_TABLE_ENTRIES; rsx++)
+			if (reg_entry_isvalid(domain->dma_table[rsx]))
+				dma_free_rt_table(domain->dma_table[rsx]);
+		break;
+	case ZPCI_TABLE_TYPE_RTX:
+		for (rtx = 0; rtx < ZPCI_TABLE_ENTRIES; rtx++)
+			if (reg_entry_isvalid(domain->dma_table[rtx]))
+				dma_free_seg_table(domain->dma_table[rtx]);
+		break;
+	default:
+		WARN_ONCE(1, "Invalid IOMMU table (%x)\n", domain->origin_type);
+		return;
+	}
 
-	dma_free_cpu_table(table);
+	dma_free_cpu_table(domain->dma_table);
 }
 
 static unsigned long *dma_alloc_page_table(gfp_t gfp)

@@ -358,7 +415,7 @@ static void s390_iommu_rcu_free_domain(struct rcu_head *head)
 {
 	struct s390_domain *s390_domain = container_of(head, struct s390_domain, rcu);
 
-	dma_cleanup_tables(s390_domain->dma_table);
+	dma_cleanup_tables(s390_domain);
 	kfree(s390_domain);
 }
 

index 42d7cc4262ca..8d8962e4fd58 100644
@@ -55,6 +55,8 @@ enum zpci_ioat_dtype {
 #define ZPCI_PT_BITS			8
 #define ZPCI_ST_SHIFT			(ZPCI_PT_BITS + PAGE_SHIFT)
 #define ZPCI_RT_SHIFT			(ZPCI_ST_SHIFT + ZPCI_TABLE_BITS)
+#define ZPCI_RS_SHIFT			(ZPCI_RT_SHIFT + ZPCI_TABLE_BITS)
+#define ZPCI_RF_SHIFT			(ZPCI_RS_SHIFT + ZPCI_TABLE_BITS)
 
 #define ZPCI_RTE_FLAG_MASK		0x3fffUL
 #define ZPCI_RTE_ADDR_MASK		(~ZPCI_RTE_FLAG_MASK)

81 lines not on this page · raw message

index 338a7381e918..46f45b136993 100644
@@ -67,6 +67,20 @@ static inline void set_pt_pfaa(unsigned long *entry, phys_addr_t pfaa)
 	*entry |= (pfaa & ZPCI_PTE_ADDR_MASK);
 }
 
+static inline void set_rf_rso(unsigned long *entry, phys_addr_t rso)
+{
+	*entry &= ZPCI_RTE_FLAG_MASK;
+	*entry |= (rso & ZPCI_RTE_ADDR_MASK);
+	*entry |= ZPCI_TABLE_TYPE_RFX;
+}
+
+static inline void set_rs_rto(unsigned long *entry, phys_addr_t rto)
+{
+	*entry &= ZPCI_RTE_FLAG_MASK;
+	*entry |= (rto & ZPCI_RTE_ADDR_MASK);
+	*entry |= ZPCI_TABLE_TYPE_RSX;
+}
+
 static inline void set_rt_sto(unsigned long *entry, phys_addr_t sto)
 {
 	*entry &= ZPCI_RTE_FLAG_MASK;

@@ -81,6 +95,22 @@ static inline void set_st_pto(unsigned long *entry, phys_addr_t pto)
 	*entry |= ZPCI_TABLE_TYPE_SX;
 }
 
+static inline void validate_rf_entry(unsigned long *entry)
+{
+	*entry &= ~ZPCI_TABLE_VALID_MASK;
+	*entry &= ~ZPCI_TABLE_OFFSET_MASK;
+	*entry |= ZPCI_TABLE_VALID;
+	*entry |= ZPCI_TABLE_LEN_RFX;
+}
+
+static inline void validate_rs_entry(unsigned long *entry)
+{
+	*entry &= ~ZPCI_TABLE_VALID_MASK;
+	*entry &= ~ZPCI_TABLE_OFFSET_MASK;
+	*entry |= ZPCI_TABLE_VALID;
+	*entry |= ZPCI_TABLE_LEN_RSX;
+}
+
 static inline void validate_rt_entry(unsigned long *entry)
 {
 	*entry &= ~ZPCI_TABLE_VALID_MASK;

@@ -286,6 +316,70 @@ static unsigned long *dma_alloc_page_table(gfp_t gfp)
 	return table;
 }
 
+static unsigned long *dma_walk_rs_table(unsigned long *rso,
+					dma_addr_t dma_addr, gfp_t gfp)
+{
+	unsigned int rsx = calc_rsx(dma_addr);
+	unsigned long old_rse, rse;
+	unsigned long *rsep, *rto;
+
+	rsep = &rso[rsx];
+	rse = READ_ONCE(*rsep);
+	if (reg_entry_isvalid(rse)) {
+		rto = get_rs_rto(rse);
+	} else {
+		rto = dma_alloc_cpu_table(gfp);
+		if (!rto)
+			return NULL;
+
+		set_rs_rto(&rse, virt_to_phys(rto));
+		validate_rs_entry(&rse);
+		entry_clr_protected(&rse);
+
+		old_rse = cmpxchg(rsep, ZPCI_TABLE_INVALID, rse);
+		if (old_rse != ZPCI_TABLE_INVALID) {
+			/* Somone else was faster, use theirs */
+			dma_free_cpu_table(rto);
+			rto = get_rs_rto(old_rse);
+		}
+	}
+	return rto;
+}
+
+static unsigned long *dma_walk_rf_table(unsigned long *rfo,
+					dma_addr_t dma_addr, gfp_t gfp)
+{
+	unsigned int rfx = calc_rfx(dma_addr);
+	unsigned long old_rfe, rfe;
+	unsigned long *rfep, *rso;
+
+	rfep = &rfo[rfx];
+	rfe = READ_ONCE(*rfep);
+	if (reg_entry_isvalid(rfe)) {
+		rso = get_rf_rso(rfe);
+	} else {
+		rso = dma_alloc_cpu_table(gfp);
+		if (!rso)
+			return NULL;
+
+		set_rf_rso(&rfe, virt_to_phys(rso));
+		validate_rf_entry(&rfe);
+		entry_clr_protected(&rfe);
+
+		old_rfe = cmpxchg(rfep, ZPCI_TABLE_INVALID, rfe);
+		if (old_rfe != ZPCI_TABLE_INVALID) {
+			/* Somone else was faster, use theirs */
+			dma_free_cpu_table(rso);
+			rso = get_rf_rso(old_rfe);
+		}
+	}
+
+	if (!rso)
+		return NULL;
+
+	return dma_walk_rs_table(rso, dma_addr, gfp);
+}
+
 static unsigned long *dma_get_seg_table_origin(unsigned long *rtep, gfp_t gfp)
 {
 	unsigned long old_rte, rte;

@@ -339,11 +433,31 @@ static unsigned long *dma_get_page_table_origin(unsigned long *step, gfp_t gfp)
 	return pto;
 }
 
-static unsigned long *dma_walk_cpu_trans(unsigned long *rto, dma_addr_t dma_addr, gfp_t gfp)
+static unsigned long *dma_walk_region_tables(struct s390_domain *domain,
+					     dma_addr_t dma_addr, gfp_t gfp)
 {
-	unsigned long *sto, *pto;
+	switch (domain->origin_type) {
+	case ZPCI_TABLE_TYPE_RFX:
+		return dma_walk_rf_table(domain->dma_table, dma_addr, gfp);
+	case ZPCI_TABLE_TYPE_RSX:
+		return dma_walk_rs_table(domain->dma_table, dma_addr, gfp);
+	case ZPCI_TABLE_TYPE_RTX:
+		return domain->dma_table;
+	default:
+		return NULL;
+	}
+}
+
+static unsigned long *dma_walk_cpu_trans(struct s390_domain *domain,
+					 dma_addr_t dma_addr, gfp_t gfp)
+{
+	unsigned long *rto, *sto, *pto;
 	unsigned int rtx, sx, px;
 
+	rto = dma_walk_region_tables(domain, dma_addr, gfp);
+	if (!rto)
+		return NULL;
+
 	rtx = calc_rtx(dma_addr);
 	sto = dma_get_seg_table_origin(&rto[rtx], gfp);
 	if (!sto)

@@ -690,8 +804,7 @@ static int s390_iommu_validate_trans(struct s390_domain *s390_domain,
 	int rc;
 
 	for (i = 0; i < nr_pages; i++) {
-		entry = dma_walk_cpu_trans(s390_domain->dma_table, dma_addr,
-					   gfp);
+		entry = dma_walk_cpu_trans(s390_domain, dma_addr, gfp);
 		if (unlikely(!entry)) {
 			rc = -ENOMEM;
 			goto undo_cpu_trans;

@@ -706,8 +819,7 @@ static int s390_iommu_validate_trans(struct s390_domain *s390_domain,
 undo_cpu_trans:
 	while (i-- > 0) {
 		dma_addr -= PAGE_SIZE;
-		entry = dma_walk_cpu_trans(s390_domain->dma_table,
-					   dma_addr, gfp);
+		entry = dma_walk_cpu_trans(s390_domain, dma_addr, gfp);
 		if (!entry)
 			break;
 		dma_update_cpu_trans(entry, 0, ZPCI_PTE_INVALID);

@@ -724,8 +836,7 @@ static int s390_iommu_invalidate_trans(struct s390_domain *s390_domain,
 	int rc = 0;
 
 	for (i = 0; i < nr_pages; i++) {
-		entry = dma_walk_cpu_trans(s390_domain->dma_table, dma_addr,
-					   GFP_ATOMIC);
+		entry = dma_walk_cpu_trans(s390_domain, dma_addr, GFP_ATOMIC);
 		if (unlikely(!entry)) {
 			rc = -EINVAL;
 			break;

index 8d8962e4fd58..d12e17201661 100644
@@ -25,6 +25,7 @@ enum zpci_ioat_dtype {
 #define ZPCI_KEY			(PAGE_DEFAULT_KEY << 5)
 
 #define ZPCI_TABLE_SIZE_RT	(1UL << 42)
+#define ZPCI_TABLE_SIZE_RS	(1UL << 53)
 
 #define ZPCI_IOTA_STO_FLAG	(ZPCI_IOTA_IOT_ENABLED | ZPCI_KEY | ZPCI_IOTA_DT_ST)
 #define ZPCI_IOTA_RTTO_FLAG	(ZPCI_IOTA_IOT_ENABLED | ZPCI_KEY | ZPCI_IOTA_DT_RT)

119 lines not on this page · raw message