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>

diff --git a/arch/s390/include/asm/pci.h b/arch/s390/include/asm/pci.h
index 8769cbf9d7cf619ab518164001ab4a2d2eec71a2..6611f798d2be6f5db9045e605259fef4994a18ef 100644 (file)
--- 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;

diff --git a/arch/s390/pci/pci_dma.c b/arch/s390/pci/pci_dma.c
index 9e5f2ecf7f25e7bab343027084e3ed51c40ef0ca..7350c8bc13a290ca362ad25b69cdb050c72ab3f4 100644 (file)
--- a/arch/s390/pci/pci_dma.c
+++ b/arch/s390/pci/pci_dma.c
@@ -257,20 +257,28 @@ static dma_addr_t dma_alloc_address(struct device *dev, int size)
        spin_lock_irqsave(&zdev->iommu_bitmap_lock, flags);
        offset = __dma_alloc_iommu(dev, zdev->next_bit, size);
        if (offset == -1) {
+               if (!zdev->tlb_refresh && !s390_iommu_strict) {
+                       /* global flush before DMA addresses are reused */
+                       if (zpci_refresh_global(zdev))
+                               goto out_error;
+
+                       bitmap_andnot(zdev->iommu_bitmap, zdev->iommu_bitmap,
+                                     zdev->lazy_bitmap, zdev->iommu_pages);
+                       bitmap_zero(zdev->lazy_bitmap, zdev->iommu_pages);
+               }
                /* wrap-around */
                offset = __dma_alloc_iommu(dev, 0, size);
-               if (offset == -1) {
-                       spin_unlock_irqrestore(&zdev->iommu_bitmap_lock, flags);
-                       return DMA_ERROR_CODE;
-               }
-               if (!zdev->tlb_refresh && !s390_iommu_strict)
-                       /* global flush after wrap-around with lazy unmap */
-                       zpci_refresh_global(zdev);
+               if (offset == -1)
+                       goto out_error;
        }
        zdev->next_bit = offset + size;
        spin_unlock_irqrestore(&zdev->iommu_bitmap_lock, flags);
 
        return zdev->start_dma + offset * PAGE_SIZE;
+
+out_error:
+       spin_unlock_irqrestore(&zdev->iommu_bitmap_lock, flags);
+       return DMA_ERROR_CODE;
 }
 
 static void dma_free_address(struct device *dev, dma_addr_t dma_addr, int size)
@@ -283,13 +291,12 @@ static void dma_free_address(struct device *dev, dma_addr_t dma_addr, int size)
        spin_lock_irqsave(&zdev->iommu_bitmap_lock, flags);
        if (!zdev->iommu_bitmap)
                goto out;
-       bitmap_clear(zdev->iommu_bitmap, offset, size);
-       /*
-        * Lazy flush for unmap: need to move next_bit to avoid address re-use
-        * until wrap-around.
-        */
-       if (!s390_iommu_strict && offset >= zdev->next_bit)
-               zdev->next_bit = offset + size;
+
+       if (zdev->tlb_refresh || s390_iommu_strict)
+               bitmap_clear(zdev->iommu_bitmap, offset, size);
+       else
+               bitmap_set(zdev->lazy_bitmap, offset, size);
+
 out:
        spin_unlock_irqrestore(&zdev->iommu_bitmap_lock, flags);
 }
@@ -557,7 +564,14 @@ int zpci_dma_init_device(struct zpci_dev *zdev)
                rc = -ENOMEM;
                goto free_dma_table;
        }
+       if (!zdev->tlb_refresh && !s390_iommu_strict) {
+               zdev->lazy_bitmap = vzalloc(zdev->iommu_pages / 8);
+               if (!zdev->lazy_bitmap) {
+                       rc = -ENOMEM;
+                       goto free_bitmap;
+               }
 
+       }
        rc = zpci_register_ioat(zdev, 0, zdev->start_dma, zdev->end_dma,
                                (u64) zdev->dma_table);
        if (rc)
@@ -567,6 +581,8 @@ int zpci_dma_init_device(struct zpci_dev *zdev)
 free_bitmap:
        vfree(zdev->iommu_bitmap);
        zdev->iommu_bitmap = NULL;
+       vfree(zdev->lazy_bitmap);
+       zdev->lazy_bitmap = NULL;
 free_dma_table:
        dma_free_cpu_table(zdev->dma_table);
        zdev->dma_table = NULL;
@@ -588,6 +604,9 @@ void zpci_dma_exit_device(struct zpci_dev *zdev)
        zdev->dma_table = NULL;
        vfree(zdev->iommu_bitmap);
        zdev->iommu_bitmap = NULL;
+       vfree(zdev->lazy_bitmap);
+       zdev->lazy_bitmap = NULL;
+
        zdev->next_bit = 0;
 }