On 12/04/2023 21:50, Marijn Suijten wrote:
On 2023-04-12 10:48:18, Abhinav Kumar wrote:
[..]
The only way to trigger this newly introduced range check is by omitting
the DSC_x constants and manually writing e.g. an out-of-range value 10
here, or setting DSC_NONE. This is only allowed for interfaces.
Correct, its just working on an implicit understanding that the indices
in the catalog
.. this sentence appears to be incomplete: what did you want to say? ..
which might still be right stick to the RM limits.
Thats why this is not bad to have.
What do you mean by "RM limits"? We have constants in the kernel that
both define the maximum number of blocks in these arrays and a
predefined set of ids that block can have. These are all used in
constant structs in the catalog, so there's nothing "software" or
SoC-specific limiting about this (except what is available in the
arrays).
[..]
I think kuogee just added this to keep it consistent with other checks
present in the RM. So I didnt see any harm with that.
Yep, that's the only reason
If he did see an issue, i will let him report that here.
If so an out-of-bounds constant was hardcoded in dpu_hw_catalog.c.
Otherwise, I dont want to spend more time discussing this bounds check
when other blocks already have it.
I'll whip up a patch to clear out the extraneous lookup (assuming there
is no other reason/dependency for it to be there...) and can follow that
up with removing these range checks of known-good values in `const
struct` fields.
Yes, please.
- Marijn
--
With best wishes
Dmitry
