dma-mapping: relax warning for per-device areas
authorRobin Murphy <robin.murphy@arm.com>
Mon, 23 Jul 2018 21:42:48 +0000 (22:42 +0100)
committerChristoph Hellwig <hch@lst.de>
Wed, 25 Jul 2018 11:32:58 +0000 (13:32 +0200)
commitd27fb99f62af7b79c542d161aa5155ed57271ddc
tree3e63ef28a310f1941ed1324683db0552566ae722
parentd72e90f33aa4709ebecc5005562f52335e106a60
dma-mapping: relax warning for per-device areas

The reasons why dma_free_attrs() should not be called from IRQ context
are not necessarily obvious and somewhat buried in the development
history, so let's start by documenting the warning itself to help anyone
who does happen to hit it and wonder what the deal is.

However, this check turns out to be slightly over-restrictive for the
way that per-device memory has been spliced into the general API, since
for that case we know that dma_declare_coherent_memory() has created an
appropriate CPU mapping for the entire area and nothing dynamic should
be happening. Given that the usage model for per-device memory is often
more akin to streaming DMA than 'real' coherent DMA (e.g. allocating and
freeing space to copy short-lived packets in and out), it is also
somewhat more reasonable for those operations to happen in IRQ handlers
for such devices.

Therefore, let's move the irqs_disabled() check down past the per-device
area hook, so that that gets a chance to resolve the request before we
reach definite "you're doing it wrong" territory.

Reported-by: Fredrik Noring <noring@nocrew.org>
Tested-by: Fredrik Noring <noring@nocrew.org>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
include/linux/dma-mapping.h