Da Zheng, le Thu 08 Jan 2009 23:10:40 +0000, a écrit : > Samuel Thibault wrote: > >Hello, > > > >Da Zheng, le Sat 03 Jan 2009 00:22:25 +0000, a écrit : > > > >>The data passed to ds_device_write() as an argument is an out-of-line > >>data in the message, and I think it's the receiver's responsibility to > >>deallocate the memory. > >> > > > >It'd be better to be sure of this :) > >I don't know that stuff so much. All I know is that gnumach doesn't > >deallocate the memory in such case, and I guess it's fine because else > >I doubt the buildds would not be able to build so much stuff without > >crashing :) > > > since gnumach doesn't deallocate the memory, someone has to do it. > actually, device_write() is used very rarely. subhurd opens almost all > devices directly, so no data goes through boot. > The only two exceptions are console and pseudo-root. > However, the data written to the console is done by > device_write_inband() where the data is transferred inside the message. > Even now I still don't know how pseudo-root is used. > > so buildds cannot prove anything in this case.
Ok so forget what I said, you know these things better than me :) Samuel
