On 2017/03/05 08:15, Tinker wrote: > Hi Stuart, > > On 2017-03-04 08:12, Stuart Henderson wrote: > > On 2017/03/04 00:30, Tinker wrote: > > > I'd like to know how the 15-25Mbps PCIe UART:s work, e.g.: > > > > > > Exar XR17V354 (4x25mbps): > .. > > > Moschip MCS9904 (4x16mbps): > .. > > > OXPCIe954: > .. > > I don't know about the Exar/Moschip, > > Got to test. > > > the OXPCIe954 doesn't work on > > OpenBSD, I had it part-working but never managed to get the baud rate > > right, the clock runs at high speed so needs a larger divisor than other > > puc(4)'s but things break somewhere. > > Hm. About how badly did it break in your experience?
The UART wasn't running at the speed set (e.g. "cu -s 19200" ran at some speed other than 19200 baud). It was consistent, i.e. a null-modem between two ports on the card worked successfully, but couldn't be used to talk to anything else. I wasn't feeling sufficiently motivated to borrow an oscilloscope to figure out the actual speed ;) > > Also you can't determine the number > > of ports from the pci id, you are supposed to read from the chip to > > figure it out, so it doesn't fit the usual pucdata.c framework. > > So the kernel needs to have a per-PCI card or chip table of number of ports, > aha. Could be done, of course, it's just a bigger change than I wanted to do without knowing if the speeds could be fixed. > What about divisors and stuff for the 15-25Mbps modes, should that work out > of the box =) Baud rate is stored in ints everywhere so that would need changing for that sort of speed for starters. Also not sure if we use the full fifos on some chips, which would certainly be wanted for higher speeds. Given the ubiquity of ethernet connections I think it would be rather niche use though.. > Btw, PCI serial devices should be superior to USB cards, shouldn't they - > USB v1 and v2 have extremely high latency, for instance. For a serial > terminal it's OK but even for an Internet connection it must suck. It doesn't seem too bad for USB UMTS adapters..
