Merge i386 and amd64 md_var.h and smp.h.

[kib]

Summary

As said in the title, the common parts are moved to x86/include/x86_smp.h and x86/include/x86_var.h.

I did not tried to fix the inthand_t and alsias_for_inthand_t mess (see types and esp. types for amd64).

[kib]

Also, the question is what copyright notice to add to the x86_smp.h. I do not like beer myself and do not want to give a bootle to phk either.

[jhb]

I've wanted to do this for md_var.h for a while. Thanks. I'm ambivalent about the license. I think you are probably stuck using it for x86_smp.h since all the subsequent changes to those files have been under that license. Also, it's not clear from arc, but are you copying these files from amd64 to x86?

sys/amd64/include/smp.h

24 ↗

(On Diff #10696)

Even though i386 doesn't use it, this one could move to the shared header since it is declared in the shared code.

sys/x86/include/x86_smp.h

16 ↗

(On Diff #10696)

Might be nice to update the comment here to x86_mp.c. Not sure if any of these are
declared in mp_machdep.c still?

[jhb]

Oh, I'm not sure what to do to fix the alias_for_inthand_t silliness. I must be a bit daft as I don't understand why setidt() can't just accept inthand_t by default? Also, inthand_t on amd64 (intr_machdep.h) is arguably "wrong", but you'd have to fix the alias. I wonder if this is just so that md_var.h doesn't require intr_machdep.h? Maybe we could move inthand_t to a more "basic" header instead?

[kib]

I am trying to answer all raised questions, I am sorry if I missed any.

1. I did not copied any file technically, when I produced the patch. I took empty x86_var.h and x86_smp.h and moved definition one-by-one. This does not have any significance on the copyright ownership.

2. I moved smp_tlb_pmap into the x86_smp.h header. My motivation for keeping it only in amd64/include/smp.h was to reduce scope.

3. Comments in x86_mp.c are updated, I claim that almost all of the functions are in x86_mp.c.

4. inthand_t is a mess. Mostly because amd64 interrupt handler calling conventions cannot be expressed in C. Might be, for amd64 we should use void ()(void) type ? But fixing this is definitely for the follow-up change.

5. inthand_t alias indeed was added because authors avoided intr_machdep.h. Looking at the intent of intr_machdep.h, it is about code for high-level interrupt handling, which is subtly unrelated to the CPU IDT stuff. I believe this explains parts of the mess. setidt() and low-level handlers do not related to intr_machdep.c.

For items 4 and 5, I will do something after the current patch is committed.

[jhb]

In D4358#92018, @kib wrote:

I am trying to answer all raised questions, I am sorry if I missed any.

1. I did not copied any file technically, when I produced the patch. I took empty x86_var.h and x86_smp.h and moved definition one-by-one. This does not have any significance on the copyright ownership.

I was asking more if you were going to use 'svn cp' for the purposes of annotate history.

2. I moved smp_tlb_pmap into the x86_smp.h header. My motivation for keeping it only in amd64/include/smp.h was to reduce scope.

Thanks.

3. Comments in x86_mp.c are updated, I claim that almost all of the functions are in x86_mp.c.

Thanks.

4. inthand_t is a mess. Mostly because amd64 interrupt handler calling conventions cannot be expressed in C. Might be, for amd64 we should use void ()(void) type ? But fixing this is definitely for the follow-up change.

5. inthand_t alias indeed was added because authors avoided intr_machdep.h. Looking at the intent of intr_machdep.h, it is about code for high-level interrupt handling, which is subtly unrelated to the CPU IDT stuff. I believe this explains parts of the mess. setidt() and low-level handlers do not related to intr_machdep.c.

Yes, I think that it is a quite reasonable split, and yes I agree that this patch shouldn't deal with it.

I think the current C prototype is wonky anyway because we never define any handlers in C anyway. In that case I think void ()(void) is more appropriate.

[kib]

In D4358#92260, @jhb wrote:

In D4358#92018, @kib wrote:

1. I did not copied any file technically, when I produced the patch. I took empty x86_var.h and x86_smp.h and moved definition one-by-one. This does not have any significance on the copyright ownership.

I was asking more if you were going to use 'svn cp' for the purposes of annotate history.

Most likely yes, I did it for the previous merges. I think that i386 version has longer history and thus is preferred.

I think the current C prototype is wonky anyway because we never define any handlers in C anyway. In that case I think void ()(void) is more appropriate.

Yes, this sounds as a good idea to change i386 to void ()(void) as well.

[emaste]

I'd prefer we end up with a standard 2 clause license if phk agrees, but am happy for you to commit this version and then make that change later.