s390/pci_dma: split dma_update_trans
Split dma_update_trans into __dma_update_trans which handles updating
the dma translation tables and __dma_purge_tlb which takes care of
purging associated entries in the dma translation lookaside buffer.
The map_sg API makes use of this split approach by calling
__dma_update_trans once per physically contiguous address range but
__dma_purge_tlb only once per dma contiguous address range.
This results in less invocations of the expensive RPCIT instruction
when using map_sg.
Signed-off-by: Sebastian Ott <[email protected]>
Reviewed-by: Gerald Schaefer <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
diff --git a/arch/s390/pci/pci_dma.c b/arch/s390/pci/pci_dma.c
index 54cb54c..9e5f2ec 100644
--- a/arch/s390/pci/pci_dma.c
+++ b/arch/s390/pci/pci_dma.c
@@ -129,12 +129,11 @@
entry_clr_protected(entry);
}
-static int dma_update_trans(struct zpci_dev *zdev, unsigned long pa,
- dma_addr_t dma_addr, size_t size, int flags)
+static int __dma_update_trans(struct zpci_dev *zdev, unsigned long pa,
+ dma_addr_t dma_addr, size_t size, int flags)
{
unsigned int nr_pages = PAGE_ALIGN(size) >> PAGE_SHIFT;
u8 *page_addr = (u8 *) (pa & PAGE_MASK);
- dma_addr_t start_dma_addr = dma_addr;
unsigned long irq_flags;
unsigned long *entry;
int i, rc = 0;
@@ -145,7 +144,7 @@
spin_lock_irqsave(&zdev->dma_table_lock, irq_flags);
if (!zdev->dma_table) {
rc = -EINVAL;
- goto no_refresh;
+ goto out_unlock;
}
for (i = 0; i < nr_pages; i++) {
@@ -159,20 +158,6 @@
dma_addr += PAGE_SIZE;
}
- /*
- * With zdev->tlb_refresh == 0, rpcit is not required to establish new
- * translations when previously invalid translation-table entries are
- * validated. With lazy unmap, it also is skipped for previously valid
- * entries, but a global rpcit is then required before any address can
- * be re-used, i.e. after each iommu bitmap wrap-around.
- */
- if (!zdev->tlb_refresh &&
- (!s390_iommu_strict ||
- ((flags & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID)))
- goto no_refresh;
-
- rc = zpci_refresh_trans((u64) zdev->fh << 32, start_dma_addr,
- nr_pages * PAGE_SIZE);
undo_cpu_trans:
if (rc && ((flags & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID)) {
flags = ZPCI_PTE_INVALID;
@@ -185,12 +170,46 @@
dma_update_cpu_trans(entry, page_addr, flags);
}
}
-
-no_refresh:
+out_unlock:
spin_unlock_irqrestore(&zdev->dma_table_lock, irq_flags);
return rc;
}
+static int __dma_purge_tlb(struct zpci_dev *zdev, dma_addr_t dma_addr,
+ size_t size, int flags)
+{
+ /*
+ * With zdev->tlb_refresh == 0, rpcit is not required to establish new
+ * translations when previously invalid translation-table entries are
+ * validated. With lazy unmap, it also is skipped for previously valid
+ * entries, but a global rpcit is then required before any address can
+ * be re-used, i.e. after each iommu bitmap wrap-around.
+ */
+ if (!zdev->tlb_refresh &&
+ (!s390_iommu_strict ||
+ ((flags & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID)))
+ return 0;
+
+ return zpci_refresh_trans((u64) zdev->fh << 32, dma_addr,
+ PAGE_ALIGN(size));
+}
+
+static int dma_update_trans(struct zpci_dev *zdev, unsigned long pa,
+ dma_addr_t dma_addr, size_t size, int flags)
+{
+ int rc;
+
+ rc = __dma_update_trans(zdev, pa, dma_addr, size, flags);
+ if (rc)
+ return rc;
+
+ rc = __dma_purge_tlb(zdev, dma_addr, size, flags);
+ if (rc && ((flags & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID))
+ __dma_update_trans(zdev, pa, dma_addr, size, ZPCI_PTE_INVALID);
+
+ return rc;
+}
+
void dma_free_seg_table(unsigned long entry)
{
unsigned long *sto = get_rt_sto(entry);
@@ -411,12 +430,16 @@
for (s = sg; dma_addr < dma_addr_base + size; s = sg_next(s)) {
pa = page_to_phys(sg_page(s)) + s->offset;
- ret = dma_update_trans(zdev, pa, dma_addr, s->length, flags);
+ ret = __dma_update_trans(zdev, pa, dma_addr, s->length, flags);
if (ret)
goto unmap;
dma_addr += s->length;
}
+ ret = __dma_purge_tlb(zdev, dma_addr_base, size, flags);
+ if (ret)
+ goto unmap;
+
*handle = dma_addr_base;
atomic64_add(size >> PAGE_SHIFT, &zdev->mapped_pages);
