• Drop src.opts.mk change.

If the BAR varies across devices, I think using pci_msix_table_bar is probably the simplest thing to do.

I think what we want to do is have a loader tunable to control this. We probably want to default to WC still, but a loader tunable makes it easy to toggle this for cases that need UC instead.

So I don't see any uses of mevent_add_disabled() in this change? If that is used in a future change it might be best to defer that to the future change and just add the new hook(s) in the backend interface for now?

• Use bsd.post.port.mk at bottom.
• Add missing svn props on intel_isakern.c.

• Use LOCALBASE for yasm.
• Make portlint mostly happy.
• Use IGNORE instead of BROKEN.
• Use PLIST_FILES instead of 1-line pkg-plist.

Does this fix anything in the base system? I thought 'ps H' worked on vmcores already? Hmm, just tried it on one and indeed it does not.

• Rebasing on today's head.

Is the intention to support remote GDB over this? If so, that seems super nifty. My only question is on the name. If this is truly restricted to only be used during panics, then panicnet is fine. OTOH, if you could use this even on a non-panicked machine (e.g. if you could configure a dedicated interface that you could attach remote GDB to or that you could send netdumps over from DDB via 'dump' before doing a continue), then maybe something like 'debugnet' is more apt?

This looks ok to me. I assume you've tested DDP and TLS?

This stuff is a bit too fragile in general, but if I fixed it, I would do a larger rototill by having MODUL_DEPEND only specify the version it is compiled against and then having modules advertise the versions they have compatible shims for (so instead of modules trying to guess a __FreeBSD_version range of kernels they are compatible with, the kernel would say which range of values it could handle modules from, and to handle things like virtual box and drm-kmod that want to use VM internals we might have a freebsd_vm "module" for which kernels advertise a smaller compatible range).

https://github.com/openssl/openssl/compare/OpenSSL_1_1_1-stable...bsdjhb:kern_tls_1_1_1 contains patches close to what has been used previously in production but updated against changes made while upstreaming. It is not against a FreeBSD source tree, but against openssl directly.

I think add a comment as @mjg requests and the existing patch is good to go. Thanks Xin Li!

In D21362#464861, @mjg wrote:

All other routines are read-write and I presume should translate automatically. Looks like everything is just redefined in terms LA_ macros anyway. If they are to mix, what's the point of having all the redefinitions?

In D21362#464856, @cem wrote:

In D21362#464855, @jhb wrote:

In D21362#464773, @mjg wrote:

I disagree with mixing up LA_ and MA_ namespaces.

This is to avoid having lookup tables in every single lc_assert method.

Thanks, I was also curious about the motivation / context for the change.

The linux lockdep API assumes LA_LOCKED semantics (meaning either a shared lock or write lock is ok). lc_assert isn't widely used in the tree, so this just hasn't been hit before.

Would it make more sense to add a LC_SHARED to lock_class::lc_flags and define a short wrapper in sys/lock.h?

static inline void
lock_assert(const struct lock_object *lo, int la_what)
{
    if ((lo->lc_flags & LC_SHARED) == 0 && (la_what & LA_LOCKED) != 0) {
        la_what &= ~LA_LOCKED;
        la_what |= LA_XLOCKED;
    }
    lo->lc_assert(lo, la_what);
}

In D21362#464773, @mjg wrote:

I disagree with mixing up LA_ and MA_ namespaces.

I would just use hselasky's approach of extending the case statements, but this is also ok.

Ping

• Use VM_ALLOC_NORMAL and VM_ALLOC_WIRED.
• Use const bool for hw_tls.

FWIW, gcc4.2.1 does get installed as /usr/bin/gcc on platforms that use it as /usr/bin/cc, and the base/gcc port that installs gcc6 as /usr/bin/cc also installs it as /usr/bin/gcc. I think gcc may be slightly different though in that autoconf scripts tend to look specifically for gcc sometimes instead of cc. I haven't run into similar configure scripts that look specifically for 'clang'. In my experience they look for 'cc' which happens to be clang.

• Bump __FreeBSD_version.

• Spelling nit noticed by arc.

• Map lockdep_is_held 1 for !INVARIANTS.

In D20992#460678, @hselasky wrote:

@jhb : OK, if you can move lockdep_is_held() under #ifdef INVARIANTS and otherwise define lockdep_is_held() as 1, I think we are in-line.

There were several other things that looked like they were also identical (e.g. several agp drivers, all the atkbdc stuff, acpi_wmi_if.m). Those can be followup commits though.

I think you did an svn merge or the like that set a crap ton of properties on lots of subdirectories? The affected files list from the original e-mail seemed to start off by listing every single directory in the source tree.

If you wanted to be super crazy you could have a NOTES.syscons file, but that is perhaps over engineering just for syscons. If you did that, I would put the screen savers in it as well and then sparc64 would keep the few 'nodevice foo_saver' it has currently.

Oh, I thought it seemed to be there from the code I read on Friday:

Ah, I had thought we would use 'include' in files.i386 and files.amd64 rather than DEFAULTS, but either way is fine with me.

OTOH, I wonder if what we might not want instead is a 'NOTES_EXTRA'. You would modify makeLINT.mk to do something like this: