Hector Martin

IOMMU, or just not having bus mastering. But yeah. Problem is none of their IOMMUs work. They don't fail closed.

Why is that? I haven't looked at how one actually programs an IOMMU. Is it a fundamental limitation or some sort of "we can't break existing devices relying on being able to read/write anywhere"?

You don't have to tell me about drivers fucking up IOMMU mappings. That's how we owned the PS4. The solution isn't forcing bounce buffers on everything, it's better software frameworks for getting this right without just dumbly copying data.

If final destination is userspace you can't. Other data may be in same page. Based on the LWN summary I posted it looks like they're doing bounce buffer mappings only for first/last partial pages...

I've seen well-written DPDK drivers that manage to do a decent job

Maybe there's something more fundamental I'm missing or being obtuse about

The only thing you need to do is make sure all your DMA allocations are page-aligned and that you never store control structures not intended for the hardware in them. Any mapped pages are treated as untrusted data subject to change at any time.