Re: block: DMA alignment of IO buffer allocated from slab

From: Bart Van Assche
Date: Mon Sep 24 2018 - 16:54:09 EST

On Mon, 2018-09-24 at 13:41 -0700, Matthew Wilcox wrote:
+AD4 On Mon, Sep 24, 2018 at 12:56:18PM -0700, Bart Van Assche wrote:
+AD4 +AD4 On Mon, 2018-09-24 at 11:57 -0700, Matthew Wilcox wrote:
+AD4 +AD4 +AD4 You're not supposed to use kmalloc memory for DMA. This is why we have
+AD4 +AD4 +AD4 dma+AF8-alloc+AF8-coherent() and friends.
+AD4 +AD4
+AD4 +AD4 Are you claiming that all drivers that use DMA should use coherent DMA only? If
+AD4 +AD4 coherent DMA is the only DMA style that should be used, why do the following
+AD4 +AD4 function pointers exist in struct dma+AF8-map+AF8-ops?
+AD4 Good job snipping the part of my reply which addressed this. Go read
+AD4 DMA-API.txt yourself. Carefully.

The snipped part did not contradict your claim that +ACI-You're not supposed to use
kmalloc memory for DMA.+ACI In the DMA-API.txt document however there are multiple
explicit statements that support allocating memory for DMA with kmalloc(). Here
is one example from the DMA-API.txt section about dma+AF8-map+AF8-single():

Not all memory regions in a machine can be mapped by this API.
Further, contiguous kernel virtual space may not be contiguous as
physical memory. Since this API does not provide any scatter/gather
capability, it will fail if the user tries to map a non-physically
contiguous piece of memory. For this reason, memory to be mapped by
this API should be obtained from sources which guarantee it to be
physically contiguous (like kmalloc).