PCI: iommu: iotlb flushing

This patch is for batching up the flushing of the IOTLB for the DMAR
implementation found in the Intel VT-d hardware.  It works by building a list
of to be flushed IOTLB entries and a bitmap list of which DMAR engine they are
from.

After either a high water mark (250 accessible via debugfs) or 10ms the list
of iova's will be reclaimed and the DMAR engines associated are IOTLB-flushed.

This approach recovers 15 to 20% of the performance lost when using the IOMMU
for my netperf udp stream benchmark with small packets.  It can be disabled
with a kernel boot parameter "intel_iommu=strict".

Its use does weaken the IOMMU protections a bit.

Signed-off-by: Mark Gross <mgross@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

1 files changed, 4 insertions, 0 deletions

diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
index e30d8fe4e4b1..f7492cd10093 100644
--- a/Documentation/kernel-parameters.txt
+++ b/Documentation/kernel-parameters.txt
@@ -847,6 +847,10 @@ and is between 256 and 4096 characters. It is defined in the file
                         than 32 bit addressing. The default is to look
                         for translation below 32 bit and if not available
                         then look in the higher range.
+                strict [Default Off]
+                        With this option on every unmap_single operation will
+                        result in a hardware IOTLB flush operation as opposed
+                        to batching them for performance.
 
         io_delay=       [X86-32,X86-64] I/O delay method
                 0x80