Re: [PATCH 07/20] iommu/amd: Remove special mapping code for dma_ops path

From: Joerg Roedel
Date: Tue Jul 12 2016 - 07:48:16 EST

Hi Robin,

On Tue, Jul 12, 2016 at 12:42:59PM +0100, Robin Murphy wrote:
> Ah, that's the angle I was missing, yes. So if, say, someone is mapping
> a page at IOVA 0x0000 while someone else is mapping a page at 0x1000,
> there could still be a race between both callers writing the non-leaf
> PTEs, but it's benign since they'd be writing identical entries anyway.
> Seems reasonable to me (I assume in a similar map vs. unmap race, the
> unmapper would just be removing the leaf entry, rather than bothering to
> check for empty tables and tear down intermediate levels, so the same
> still applies).

The non-leaf PTE setup code checks for races with cmpxchg, so we are on
the safe side there. Two threads would write different entries there,
because they are allocating differnt sub-pages, but as I said, this is
checked for using cmpxchg. On the PTE level this problem does not exist.