Discussion:
[OpenRISC] l.sys and EEAR.
Matt Thomas
2014-09-05 16:42:06 UTC
Permalink
Having a syscall number is l.sys isn't that useful since you need to fetch the instruction to read it.
If l.sys could place it in the EEAR SPR so the syscall handler could fetch it would allow skipping of
loading the syscall number in a register.
Sébastien Bourdeauducq
2014-09-05 16:53:48 UTC
Permalink
Post by Matt Thomas
Having a syscall number is l.sys isn't that useful since you need to fetch the instruction to read it.
If l.sys could place it in the EEAR SPR so the syscall handler could fetch it would allow skipping of
loading the syscall number in a register.
Yes. And the exception vector addresses could use some cleanup too...
Stefan Kristiansson
2014-09-05 17:45:23 UTC
Permalink
Post by Matt Thomas
Having a syscall number is l.sys isn't that useful since you need to fetch the instruction to read it.
If l.sys could place it in the EEAR SPR so the syscall handler could fetch it would allow skipping of
loading the syscall number in a register.
The immediate in the l.sys isn't very useful, no.
I suspect it was thrown in there since the space was unused anyway,
but afaik no-one has ever used that.
Storing the immediate into EEAR SPR is an interesting idea, and while
that is worth considering to add to the arch spec (it doesn't break
anything), depending on it will of course limit you to implementations
post that version of the spec.

Stefan
Sébastien Bourdeauducq
2014-09-06 03:17:12 UTC
Permalink
that is worth considering to add to the arch spec, depending on it
will of course limit you to implementations post that version of the
spec.
Do people really care so much about that? As far as I'm concerned, what
bothers me about OpenRISC is its bloat and its bad technical choices
such as those just discussed - not the lack of bureaucracy and red tape.
I wonder if others would agree.

Sébastien

Loading...