I'm okay with the idea but I'm a little worried that it changes the semantics of vm_page_insert / vm_page_insert_after, which I just preserved with the insert_lookup patch. Those two procedures have several callers. However, several of the callers appear to be broken (lacking error checks), so maybe we should audit them all anyway.

After some experiments, I think that [[noreturn]] is not a solution that either clang or g++ will accept.

In D46014#1049503, @jhb wrote:

In D46014#1048959, @rlibby wrote:

Also, I think this kind of pattern is behind many of our -Warray-bounds
violations too, where we allocate a part of a structure and then rely on
code referring to only those parts. Especially in areas like cam. This
one is a little different in that we apparently also actually access
beyond the short allocation.

Well, some of those are due to using [0] arrays instead of the more modern syntax for variable length arrays.

Hmm, I've been avoiding addressing warnings in contrib...

I guess you can also consider if you want to make the malloc spelling more conventional. Don't need the casts, and can now do sizeof(*asyreq). I'm ambivalent, but in case you hadn't considered it.

I don't know this code but I'm assuming that we don't really care about
the overhead here or performance generally for this ioctl. In other
words, we don't need a longer term fix to do a saner amount of copying,
correct? It seems that we never actually use asyreq->data in the ioctl,
so we're now allocating and passing around an extra 2k because of a
weird definition?

I haven't looked at swap_pager.c yet. I'll get back to that hopefully tonight.

Any more input on this one?

kib: style fixup

In D45918#1046830, @kib wrote:

It seems that db_error() is the only caller of kdb_reenter_silent(). And then it might be better to panic in kdb_reenter_silent() instead, I am not sure. [To make kdb_reenter_silent() more strict]

In D45891#1046815, @arichardson wrote:

No sure how much the builtins optimization matters here but if it does the other workaround would be:

void* memcpy_for_plt(void*, const void*, size_t) asm("memcpy")

And then call that.

kib: comment the cflag

kib: just disable all builtins

In D45891#1046235, @kib wrote:

Should we disable all built-ins for the file, for the same reasoning?

Yeah, we should use the value from sx_try_lock, and we'd like to enforce that it's checked too I guess.

imp: sx_xlock too

In D45875#1045918, @imp wrote:

Why didn't sx_lock() complain?

In D45826#1045245, @kib wrote:

In D45826#1044991, @rlibby wrote:

By the way, I thought about also making these strict-aliasing safe with a union. It's straightforward and also gets rid of the casts. However, since we explicitly build with -fno-strict-aliasing, there's not much motivation. But in case you'd prefer that, I can tweak these patches.

Why would a special treat for aliasing needed? There is the "memory" clobber.

By the way, I thought about also making these strict-aliasing safe with a union. It's straightforward and also gets rid of the casts. However, since we explicitly build with -fno-strict-aliasing, there's not much motivation. But in case you'd prefer that, I can tweak these patches.

In D45825#1044906, @imp wrote:

Ok. So non c11 env?

In D45825#1044894, @imp wrote:

Seems fine... what compilers / build env hits this?

diff -u1000 sys/{i386,amd64}/include/pcpu.h for comparison to D45826. Ignore the #if 0 for testing.

Test patch used to examine codegen:

Filed https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=280039