s390/pci_dma: improve lazy flush for unmap

Lazy unmap (defer tlb flush after unmap until dma address reuse) can
greatly reduce the number of RPCIT instructions in the best case. In
reality we are often far away from the best case scenario because our
implementation suffers from the following problem:

To create dma addresses we maintain an iommu bitmap and a pointer into
that bitmap to mark the start of the next search. That pointer moves from
the start to the end of that bitmap and we issue a global tlb flush
once that pointer wraps around. To prevent address reuse before we issue
the tlb flush we even have to move the next pointer during unmaps - when
clearing a bit > next. This could lead to a situation where we only use
the rear part of that bitmap and issue more tlb flushes than expected.

To fix this we no longer clear bits during unmap but maintain a 2nd
bitmap which we use to mark addresses that can't be reused until we issue
the global tlb flush after wrap around.

Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Reviewed-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

1 files changed, 1 insertions, 0 deletions

diff --git a/arch/s390/include/asm/pci.h b/arch/s390/include/asm/pci.h
index 8769cbf..6611f79 100644
--- a/arch/s390/include/asm/pci.h
+++ b/arch/s390/include/asm/pci.h
@@ -118,6 +118,7 @@ struct zpci_dev {
 
 	spinlock_t	iommu_bitmap_lock;
 	unsigned long	*iommu_bitmap;
+	unsigned long	*lazy_bitmap;
 	unsigned long	iommu_size;
 	unsigned long	iommu_pages;
 	unsigned int	next_bit;