On 29/10/2026 20:45, Olivier Valentin wrote:
Le 29 octobre 2026 21:28:14 GMT+01:00, Michael Kelly <[email protected]> a
écrit :
I'll see if portions of a modern version of Linux pci/probe.c can be merged or
if there is an alternative simpler workaround (assuming that the above is the
cause, which does seem likely).
Last week, when I looked into this code, it seemed that the test for 64bits
BARs support depends on the size of a pointer type. So I was suspecting that
ddekit was not configured correctly in the case of x86_64 (from the top of my
head, it looks like CONFIG_...PCI_64BITS).
But I could not find a single place where this config can be set.
As you say, maybe it was fixed in a later version ?
Adding CONFIG_PHYS_ADDR_T_64BIT to the
libdde-linux26/include/linux/autoconf.h fixes this particular issue. I
have tested a build of libdde-linux26 and netdde with this configuration
option on hurd-amd64, hurd-i386 and hurd-i386-pae. Supporting the 64 bit
physical address using a hurd-i386 kernel has no value but since the 32
bit netdde can be run on hurd-i386-pae, I used the
CONFIG_PHYS_ADDR_T_64BIT option throughout.
I can see that there is a risk some netdde drivers might now fail and I
have only performed minimal testing using the e1000 driver in all cases.
I could provide a hurd/debian/patches patch, but I can't find those in
the hurd git repository. Where are they, please?
Cheers,
Mike.
