From: "Theo de Raadt" <deraadt@openbsd.org>
Subject: Re: bge 64 bit dma
To: Mark Kettenis <mark.kettenis@xs4all.nl>
Cc: Alexander Bluhm <bluhm@openbsd.org>, tech@openbsd.org
Date: Fri, 27 Feb 2026 18:25:39 -0700

Mark Kettenis <mark.kettenis@xs4all.nl> wrote:

> > Date: Fri, 27 Feb 2026 23:47:57 +0100
> > From: Alexander Bluhm <bluhm@openbsd.org>
> > 
> > Hi,
> > 
> > Brad sent me this diff.  I have tested it on
> > 
> > bge0 at pci12 dev 0 function 0 "Broadcom BCM5720" rev 0x00, BCM5720 A0 (0x5720000): msi, address 04:32:01:f1:4e:7c
> > 
> > and Hrvoje here
> > 
> > bge1 at pci14 dev 0 function 1 "Broadcom BCM5720" rev 0x00, BCM5720 A0 (0x5720000), APE firmware NCSI 1.5.61.0: msi, address d0:94:66:5b:66:0e
> > 
> > Is that enough test coverage to enable 64 bit dma for bge(4)?
> 
> According to the Linux tg3 driver, some variants only do 32-bit DMA or
> have an integrated PCI-PCI bridge that can only do 40-bit DMA.  It is
> not clear to me how those map onto PCIe vs non-PCIe.

I agree, but I think there is a stronger trend:

The broadcom product space is polluted by very strange MIPS cpu variants
with incompplete address reachability, acting as dma engines and address
space translators.  The Broadcom "undocumented address-space-limitations"
blast radius is probably large.   We need to be careful.