On Sun, Feb 02, 2020 at 11:46:44PM -0800, Matthew Wilcox wrote:

quoted

gives the hardware access to completely unrelated memory.) Instead,
they get pages from the page allocator, and these pages may e.g. be
allocated from the DMA, DMA32 or NORMAL zones depending on the
restrictions imposed by hardware. So I think the usercopy restriction
only affects a few oddball drivers (like this s390 stuff), which is
why you're not seeing more bug reports caused by this.

Getting pages from the page allocator is true for dma_alloc_coherent()
and friends.

dma_alloc_coherent also has a few other memory sources than the page
allocator..

But it's not true for streaming DMA mappings (dma_map_*)
for which the memory usually comes from kmalloc().  If this is something
we want to fix (and I have an awful feeling we're going to regret it
if we say "no, we trust the hardware"), we're going to have to come up
with a new memory allocation API for these cases.  Or bounce bugger the
memory for devices we don't trust.

There aren't too many places that use slab allocations for streaming
mappings, and then do usercopies into them.  But I vaguely remember
some usb code getting the annotations for that a while ago.

But if you don't trust your hardware you will need to use IOMMUs and
page aligned memory, or IOMMUs plus bounce buffering for the kmalloc
allocations (we've recently grown code to do that for the intel-iommu
driver, which needs to be lifted into the generic IOMMU code).