On Thu, Jun 26, 2025 at 09:24:48PM +0300, Jarkko Sakkinen wrote:
> > At least I would add the class and drop the tpm_buf_destroy() as one
> > patch, and another would be to cleanup any empty gotos.
> >
> > Also, I think the style guide for cleanup.h is to not use the
> > variable block, so it should be more like:
> >
> > CLASS(tpm_buf, buf)();
> > if (!tpm_buf)
> > return -ENOMEM;
> >
> > AFAICT, but that seems to be some kind of tribal knowledge.
>
> This was improved in v2 :-) If you have some proposal how you'd
> liked that version to be splitted, please give feedback.
After a bit of thought, II could split v2 e.g., into to the following
list of patches (a draft, along the lines):
1. Prepare internals for API changes.
2. Implement tpm_buf_alloc().
3. Implement CLASS_TPM_BUF() macro.
4. Changes for tpm{1,2}-cmd.c.
6. Changes for tpm2-sessions.c.
7. Changes for tpm2-space.c.
8. Changes for trusted_tpm{1,2}.c
9. Remove stuff left w/o a call site.
It's pretty good exercise for v2 actually as it is already somewhat
functional code. By doing this split this update will get already
reasonably well verified.
I should also probably emphasize the motivation better in the next
version. Especially with multiple tpm_buf instances in the same
function scope, things do something are messy to backtrack. In addition,
this complexity might cap the motivation for someone to contribute a
useful feature.
I don't really have even followed Linus' opinions in this topic per se
I personally just think that since I have a measured argument for this.
I got with that and talk with Linus if he wants to bring it up :-)
BR, Jarkko
