s390/pci: use lock-free I/O translation updates

I/O translation tables on s390 use 8 byte page table entries and tables
which are allocated lazily but only freed when the entire I/O
translation table is torn down. Also each IOVA can at any time only
translate to one physical address Furthermore I/O table accesses by the
IOMMU hardware are cache coherent. With a bit of care we can thus use
atomic updates to manipulate the translation table without having to use
a global lock at all. This is done analogous to the existing I/O
translation table handling code used on Intel and AMD x86 systems.

Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
Link: https://lore.kernel.org/r/20221109142903.4080275-6-schnelle@linux.ibm.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>

2 files changed, 45 insertions, 30 deletions

diff --git a/arch/s390/include/asm/pci.h b/arch/s390/include/asm/pci.h
index e4c3e4e04d30..b248694e0024 100644
--- a/arch/s390/include/asm/pci.h
+++ b/arch/s390/include/asm/pci.h
@@ -157,7 +157,6 @@ struct zpci_dev {
 
 	/* DMA stuff */
 	unsigned long	*dma_table;
-	spinlock_t	dma_table_lock;
 	int		tlb_refresh;
 
 	spinlock_t	iommu_bitmap_lock;
diff --git a/arch/s390/pci/pci_dma.c b/arch/s390/pci/pci_dma.c
index dee825ee7305..ea478d11fbd1 100644
--- a/arch/s390/pci/pci_dma.c
+++ b/arch/s390/pci/pci_dma.c
@@ -63,37 +63,55 @@ static void dma_free_page_table(void *table)
 	kmem_cache_free(dma_page_table_cache, table);
 }
 
-static unsigned long *dma_get_seg_table_origin(unsigned long *entry)
+static unsigned long *dma_get_seg_table_origin(unsigned long *rtep)
 {
+	unsigned long old_rte, rte;
 	unsigned long *sto;
 
-	if (reg_entry_isvalid(*entry))
-		sto = get_rt_sto(*entry);
-	else {
+	rte = READ_ONCE(*rtep);
+	if (reg_entry_isvalid(rte)) {
+		sto = get_rt_sto(rte);
+	} else {
 		sto = dma_alloc_cpu_table();
 		if (!sto)
 			return NULL;
 
-		set_rt_sto(entry, virt_to_phys(sto));
-		validate_rt_entry(entry);
-		entry_clr_protected(entry);
+		set_rt_sto(&rte, virt_to_phys(sto));
+		validate_rt_entry(&rte);
+		entry_clr_protected(&rte);
+
+		old_rte = cmpxchg(rtep, ZPCI_TABLE_INVALID, rte);
+		if (old_rte != ZPCI_TABLE_INVALID) {
+			/* Somone else was faster, use theirs */
+			dma_free_cpu_table(sto);
+			sto = get_rt_sto(old_rte);
+		}
 	}
 	return sto;
 }
 
-static unsigned long *dma_get_page_table_origin(unsigned long *entry)
+static unsigned long *dma_get_page_table_origin(unsigned long *step)
 {
+	unsigned long old_ste, ste;
 	unsigned long *pto;
 
-	if (reg_entry_isvalid(*entry))
-		pto = get_st_pto(*entry);
-	else {
+	ste = READ_ONCE(*step);
+	if (reg_entry_isvalid(ste)) {
+		pto = get_st_pto(ste);
+	} else {
 		pto = dma_alloc_page_table();
 		if (!pto)
 			return NULL;
-		set_st_pto(entry, virt_to_phys(pto));
-		validate_st_entry(entry);
-		entry_clr_protected(entry);
+		set_st_pto(&ste, virt_to_phys(pto));
+		validate_st_entry(&ste);
+		entry_clr_protected(&ste);
+
+		old_ste = cmpxchg(step, ZPCI_TABLE_INVALID, ste);
+		if (old_ste != ZPCI_TABLE_INVALID) {
+			/* Somone else was faster, use theirs */
+			dma_free_page_table(pto);
+			pto = get_st_pto(old_ste);
+		}
 	}
 	return pto;
 }
@@ -117,19 +135,24 @@ unsigned long *dma_walk_cpu_trans(unsigned long *rto, dma_addr_t dma_addr)
 	return &pto[px];
 }
 
-void dma_update_cpu_trans(unsigned long *entry, phys_addr_t page_addr, int flags)
+void dma_update_cpu_trans(unsigned long *ptep, phys_addr_t page_addr, int flags)
 {
+	unsigned long pte;
+
+	pte = READ_ONCE(*ptep);
 	if (flags & ZPCI_PTE_INVALID) {
-		invalidate_pt_entry(entry);
+		invalidate_pt_entry(&pte);
 	} else {
-		set_pt_pfaa(entry, page_addr);
-		validate_pt_entry(entry);
+		set_pt_pfaa(&pte, page_addr);
+		validate_pt_entry(&pte);
 	}
 
 	if (flags & ZPCI_TABLE_PROTECTED)
-		entry_set_protected(entry);
+		entry_set_protected(&pte);
 	else
-		entry_clr_protected(entry);
+		entry_clr_protected(&pte);
+
+	xchg(ptep, pte);
 }
 
 static int __dma_update_trans(struct zpci_dev *zdev, phys_addr_t pa,
@@ -137,18 +160,14 @@ static int __dma_update_trans(struct zpci_dev *zdev, phys_addr_t pa,
 {
 	unsigned int nr_pages = PAGE_ALIGN(size) >> PAGE_SHIFT;
 	phys_addr_t page_addr = (pa & PAGE_MASK);
-	unsigned long irq_flags;
 	unsigned long *entry;
 	int i, rc = 0;
 
 	if (!nr_pages)
 		return -EINVAL;
 
-	spin_lock_irqsave(&zdev->dma_table_lock, irq_flags);
-	if (!zdev->dma_table) {
-		rc = -EINVAL;
-		goto out_unlock;
-	}
+	if (!zdev->dma_table)
+		return -EINVAL;
 
 	for (i = 0; i < nr_pages; i++) {
 		entry = dma_walk_cpu_trans(zdev->dma_table, dma_addr);
@@ -173,8 +192,6 @@ undo_cpu_trans:
 			dma_update_cpu_trans(entry, page_addr, flags);
 		}
 	}
-out_unlock:
-	spin_unlock_irqrestore(&zdev->dma_table_lock, irq_flags);
 	return rc;
 }
 
@@ -558,7 +575,6 @@ int zpci_dma_init_device(struct zpci_dev *zdev)
 	WARN_ON(zdev->s390_domain);
 
 	spin_lock_init(&zdev->iommu_bitmap_lock);
-	spin_lock_init(&zdev->dma_table_lock);
 
 	zdev->dma_table = dma_alloc_cpu_table();
 	if (!zdev->dma_table) {