DragonFly kernel List (threaded) for 2004-08
Re: PCI Express
On Wed, Aug 25, 2004 at 03:13:15PM +0200, Jeroen Ruigrok/asmodai wrote:
> I doubt any drivers or lowlevel routines need to be rewritten to accomodate
> for the shared -> point-to-point change in the specification.
Correct. The mandate for PCIe was backwards compatibility for
software. For example, I was able to use a new express device with
relatively few changes compared to the PCI-X driver. At this point, my
belief is that the bigest changes are for
- there are a handful of express registers that allow the kernel or
drivers to tune the size of requests going across the links, detect
link errors, retrain the links, etc. none of these are mandatory to
use but can be quite useful.
- config space is now 4096 bytes as opposed to 256. most of the
express control and status structures are located above offset
0x100. to access these registers, there is a memory range that
decodes memory access to config packets with the extra register
- MSI is now mandatory for devices, although legacy INTx support
exists to help with the transition to express.
- Active State Power Management (ASPM). I haven't played with this and
don't know how many first generation devices are going to support it.
Agilent Technologies, Storage Area Networking