Re: [PATCH 5.10 11/38] swiotlb: rework "fix info leak with DMA_FROM_DEVICE"

[Greg Kroah-Hartman]

On Sat, Mar 26, 2022 at 11:41:17AM -0700, Linus Torvalds wrote:
> On Sat, Mar 26, 2022 at 3:18 AM Greg Kroah-Hartman
> <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:"
> >
> > Yes, I've been watching that thread.  This change is already in 5.15 and
> > 5.16 kernels, and does solve one known security issue, so it's a tough
> > call.
> 
> If you're following that thread, you'll have seen that I've reverted
> it, and I actually think the security argument was bogus - the whole
> commit was due to a misunderstanding of the actual direction of the
> data transfer.

I see that now, thanks.

But why did you just revert that commit, and not the previous one (i.e.
the one that this one "fixes")?  Shouldn't ddbd89deb7d3 ("swiotlb: fix
info leak with DMA_FROM_DEVICE") also be dropped?

I'm going to drop both from the 5.4 and 5.10 stable queues now, and add
your revert, but I think your tree also needs the original swiotlb fix
commit reverted to get back to a "known good" state.

thanks,

greg k-h