On Fri, Jun 25, 2004 at 09:38:44 -0500, Timur Tabi wrote: > Jan Hudec wrote: > > >Are you sure? The sys_mlock manipulates userspace memory. But a driver > >does not have any userspace memory to manipulate. It only has kernel > >memory and that is not swappable. > > We use sys_mlock() to lock user-space memory. We need to create a list > of physical addresses for every page in the app's buffer. This buffer > needs to be locked down and then its contents DMA'd to our hardware. No. That's wrong. You need to lock_kiobuf, just like direct read and write do. > >Because that function is a syscall. You should call it as a syscall. But > >since it only affects userspace memory, the userspace should call it. > >Really. > > Userspace can't call it because only root can call sys_mlock() and we > can't require that the calling process be root. > > >Could you please tell us what you want to do? It seems quite likely that > >there is a better solution than trying to call sys_mlock. > > I agree, but I haven't been able to find it yet. Look at the kiobuf stuff. ------------------------------------------------------------------------------- Jan 'Bulb' Hudec <bulb@ucw.cz>
Attachment:
signature.asc
Description: Digital signature
