Re: network devices don't handle pci_dma_mapping_error()'s

On Wed, 06 Dec 2006 23:24:59 -0800 (PST)
David Miller [off-list ref] wrote:

From: "Amit S. Kale" <redacted>
Date: Thu, 7 Dec 2006 11:55:22 +0530

quoted

We can let a driver handle dma mapping errors using these->

1.Reduce the size of a receive ring. This will free some possibly remapped 
memory, reducing pressure on iommu. We also need to printk a message so that 
a user knows the reason why receive ring was shrunk. Growing it when iommu 
pressure goes down will result in a ping-pong.
2. Force processing of receive and transmit ring. This will ensure that the 
buffers processed by hardware are freed, reducing iommu pressure.

3. If we need to do (1) and (2) a predefined number of times (say 20), stop 
the queue. Stopping the queue in general will cause a ping-pong, so it should 
be avoided as far as possible.

This scheme assumes the networking card is the culprit.  In many
workloads it will not be and these efforts will be in vain and perhaps
even make the situation worse.  There's not reason to run the RX and
TX queues, and even shrink them, when the FC controller has most of
the IOMMU entires tied up.

That's why users needs to queue up and get feedback when IOMMU space
is made available.

Looking at other subsystems, the disk code seems to return I/O errors
if dma mapping fails. Perhaps this discussion needs to move off to lkml
or the platform lists.