Hi, On Mon, 20 Jan 2003, Jos Hulzink wrote:
> On Mon, 20 Jan 2003, Fabio Alemagna wrote: > > > On Mon, 20 Jan 2003, Brian S. Julin wrote: > > > > Stupid question from an outsider: couldn't it be possible to make the > > application still run by making it use an offscreen buffer while not > > "visible" because of VT switching? It would be really annoying, imho, if > > the application stopped altogether... > > Yeah, nice... My 32 MB Matrox deals with 4 devices, which means you need > 4x32 MB = 128 MB offscreen for that one alone. Glad I don't own a 128 MB > ATI 9700 ... my box got only 256 MB... > > And now serious: Nice idea, but doesn't work. As you can see above, fake > framebuffers are no good but for text modes. If we really want to make a > chance to make it in any kernel, we should come up with something better. > > Jos > IMO a signal is enough. The signal tells userspace that it had better stop writing to it's mmap'd resources or the kernel will be very unhappy. If userspace feels the need to continue writing somewhere it can malloc some memory and swap pointers in the signal handlers. This way, KGI doesn't have to worry about and memory management and it's left to userspace to decide what to do and how much of a memory hog to be :P Of course, this only works for apps that write directly to the framebuffer but it's a start :) Paul
