Revert "vring: Force use of DMA API for ARM-based systems with legacy devices"
authorMichael S. Tsirkin <mst@redhat.com>
Fri, 3 Feb 2017 03:43:52 +0000 (05:43 +0200)
committerMichael S. Tsirkin <mst@redhat.com>
Fri, 3 Feb 2017 21:38:50 +0000 (23:38 +0200)
commit0d5415b489f68b58e1983a53793d25d53098ed4b
tree9509a05b5c480b84214714698cffb85943be8124
parent566cf877a1fcb6d6dc0126b076aad062054c2637
Revert "vring: Force use of DMA API for ARM-based systems with legacy devices"

This reverts commit c7070619f3408d9a0dffbed9149e6f00479cf43b.

This has been shown to regress on some ARM systems:

by forcing on DMA API usage for ARM systems, we have inadvertently
kicked open a hornets' nest in terms of cache-coherency. Namely that
unless the virtio device is explicitly described as capable of coherent
DMA by firmware, the DMA APIs on ARM and other DT-based platforms will
assume it is non-coherent. This turns out to cause a big problem for the
likes of QEMU and kvmtool, which generate virtio-mmio devices in their
guest DTs but neglect to add the often-overlooked "dma-coherent"
property; as a result, we end up with the guest making non-cacheable
accesses to the vring, the host doing so cacheably, both talking past
each other and things going horribly wrong.

We are working on a safer work-around.

Fixes: c7070619f340 ("vring: Force use of DMA API for ARM-based systems with legacy devices")
Reported-by: Robin Murphy <robin.murphy@arm.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
drivers/virtio/virtio_ring.c