> in comp.os.linux.hardware David Parsons recently wrote:
>
...
> > I keep a 1.2.13 kernel (with st.c hacked to support 256k blocksizes)
> > around just to read tapes from SGIs.
>
> any chance that this can be fixed/patched for 2.0.X kernels (doesn't need
> to go into main stream if nasty/ugly) ?
>
If you get an Alpha or Sparc64 (?) you can use a 256 kB buffer ;-)
Last year Joerg Weule emailed me his solution that used a statically
allocated buffer to bypass the limit. I don't have this solution as a
patch and this is not a solution for the mainstream kernel (but shows that
large blocks are possible).
> if not, is using >128K tape blocks possible with recent (or future) 2.1.X ?
>
It is possible that this will change in future. The tape driver can be
modified to use scatter/gather and this allows larger buffers. This also
increases the chances to successfully allocate a buffer when the kernel
memory has been fragmented. However, there are a lot of small details that
have to be handled correctly when the driver is modified and this takes
some time.
Kai