Joshua Branson, le ven. 09 nov. 2018 11:59:31 -0500, a ecrit: > > It's correct, but could be improved: I'd say rather take the example of > > glibc's send(), which is an RPC handled by pfinet, which uses a > > device_write RPC to actually emit an Ethernet frame, which is handled by > > netdde, which pushes the hardware, and gets an interrupt from GNU Mach > > when that's done. > > I think the new attached patch shows that, but I'm not an expert.
It's more interesting, yet not as good as it could :) > + --------------------------------------- > + | \ Hurd Servers | > + | | | > + | | auth and other servers| > + | | | > + | pfinet -> device_write RPC | > + | \ | | > + | \ | | > + | \ | | > + -------------------------------------- > + \ / > + \ netdde netdde is actually one of the hurd servers. I'd say either drop the Hurd Servers frame and use frames around pfinet and netdde, or keep the Hurd Servers frame, but put frames around pfinet and netdde. Otherwise it's misguiding, the reader could think that the "Hurd Servers" is just one process. pfinet itself doesn't talk with GNU Mach, it really only uses device_write implemented by netdde, which pushes to the hardware (after asking GNU Mach for permission) Samuel
