[LinuxBIOS] [PATCH] VIA VT8237R support
rminnich at gmail.com
Sun Oct 28 18:21:42 CET 2007
On 10/28/07, Stefan Reinauer <stepan at coresystems.de> wrote:
> * ron minnich <rminnich at gmail.com> [071027 19:15]:
> > one thing though. We can NEVER put a patch in that sets bus master.
> > Any BIOS that does this is broken, and any OS that requires it is
> > broken.
> Which standard says this?
none. But this is a frequent discussion on this list over the years.
It's very dangerous to enable bus mastering on a device and enable it
as well. You are inviting the device to do I/O to your memory at that
point. If it is after a reset, or some other unclean shutdown (e.g.
panic) and the device is in an intermediate state, anything can
happen. That is why (e.g.) several kernels (including LInux) had to be
re-engineered a bit to properly shut down PCI devices on modunload.
This problem became particularly important when kexec came along.
Our conclusion from a few years back is that we could and would (and
should) let the OS enable bus master.
More information about the coreboot