On Sun, 2005-02-20 at 14:54 -0700, Jamie Painter wrote: > > Here's a second try at the patch that fixes ax88172 receive > performance in full duplex mode. > > Dealing with the async read/write_cmd calls looked to me to be messy, > with four calls needed to read the link partner capabilities then set > the medium mode. I instead opted for a kevent, so I could use the > blocking calls in process context. If this is a bad decision for > some reason, please educate me -- I'm not experienced with linux > driver hacking. > > I'd also like to point out that the usbnet.c patch for AX88772 support > recently applied also issues AX_CMD_WRITE_MEDIUM_MODE. It does it in > the bind function and does not take into account the state of the link > (10hd,100fd, etc). It's possible the same performance issue I > observed with the ax88172 applies there too. It might be worth > testing link changes between fd and hd to make sure it behaves well. > I don't have a device with an AX88772 or I'd try it myself. >
I like the approach of using the kevent. It stays adapter neutral and keeps the code additions down. I don't see anything wrong with it myself, maybe David B would know if there are any concerns with it's use. If that is an acceptable solution, that would really help us out with the AX88178 chip and it's medium mode handling. As for the AX88772, it would be very similar to the 172, just poking a few different registers, but we could definitely move it into the reset function and pull it out of bind. -- David Hollis <[EMAIL PROTECTED]>
signature.asc
Description: This is a digitally signed message part
