illumos gate

In addition to the checks I've done to confirm that objects from a gcc 10 primary build are now much closer to those from gcc 7, I ran a wsdiff comparing before and after the associated patch with gcc7 (as that's the current primary compiler).

usr/sbin/iasl
usr/sbin/acpixtract
usr/sbin/acpidump
etc/versions/build
etc/motd
platform/i86pc/kernel/amd64/unix
platform/i86xpv/kernel/amd64/unix

These changes are a result of the git commit hash changing between builds (since I cherry picked the change).

usr/lib/smbsrv/amd64/libfksmbsrv.so.1
usr/lib/smbsrv/libfksmbsrv.so.1
usr/lib/smbfs/fksmbcl
usr/lib/smbfs/amd64/libfknsmb.so.1
usr/lib/smbfs/libfknsmb.so.1
usr/lib/smbfs/amd64/libfksmbfs.so.1
usr/lib/smbfs/libfksmbfs.so.1

opt/os-tests/tests/cores/amd64/libdumper.so.1
opt/os-tests/tests/cores/dumper.64
opt/os-tests/tests/cores/dumper.32
opt/os-tests/tests/cores/libdumper.so.1

Changes in .debug_{str,info} sections due to differing gcc options.

usr/bin/raidz_test

The ilog2() function has changed as follows. The change moved this function from raidz_test.h to raidz_test.c and removes the inline keyword (it was never inlined due to recursion). This has obviously changed the code generation in some way, but it still looks right, just slightly less optimised for the terminal case. I ran some ZFS tests that use the raidz_test binary and they succeeded.

ilog2()
-    ilog2:                 31 c0              xorl   %eax,%eax
-    ilog2+0x2:             48 83 ff 01        cmpq   $0x1,%rdi
-    ilog2+0x6:             77 08              ja     +0x8      <ilog2+0x10>
-    ilog2+0x8:             f3 c3              repz ret
-    ilog2+0xa:             66 0f 1f 44 00 00  nopw   0x0(%rax,%rax)
-    ilog2+0x10:            48 89 f8           movq   %rdi,%rax
-    ilog2+0x13:            48 d1 e8           shrq   $0x1,%rax
-    ilog2+0x16:            48 83 f8 01        cmpq   $0x1,%rax
-    ilog2+0x1a:            74 ec              je     -0x14     <ilog2+0x8>
-    ilog2+0x1c:            55                 pushq  %rbp
-    ilog2+0x1d:            48 c1 ef 02        shrq   $0x2,%rdi
-    ilog2+0x21:            48 89 e5           movq   %rsp,%rbp
-    ilog2+0x24:            e8 d7 ff ff ff     call   -0x29     <ilog2>
-    ilog2+0x29:            5d                 popq   %rbp
-    ilog2+0x2a:            48 83 c0 02        addq   $0x2,%rax
-    ilog2+0x2e:            c3                 ret
-    .text+0x1cf:           90                 nop
+    ilog2:                 48 83 ff 01        cmpq   $0x1,%rdi
+    ilog2+0x4:             77 0a              ja     +0xa      <ilog2+0x10>
+    ilog2+0x6:             31 c0              xorl   %eax,%eax
+    ilog2+0x8:             c3                 ret
+    ilog2+0x9:             0f 1f 80 00 00 00  nopl   0x0(%rax)
+                           00
+    ilog2+0x10:            55                 pushq  %rbp
+    ilog2+0x11:            48 d1 ef           shrq   $0x1,%rdi
+    ilog2+0x14:            48 89 e5           movq   %rsp,%rbp
+    ilog2+0x17:            e8 e4 ff ff ff     call   -0x1c     <ilog2>
+    ilog2+0x1c:            5d                 popq   %rbp
+    ilog2+0x1d:            48 ff c0           incq   %rax
+    ilog2+0x20:            c3                 ret

lib/libld.so.4
lib/amd64/libld.so.4

The is_name_cmp() function is no longer being inlined by gcc7.
gcc10 will never inline this and reports inlining failed in call to 'is_name_cmp': call is unlikely and code size would grow

lib/libsqlite-native.o

This is another object which is built and left containing DWARF. The wsdiff differences are due to the differing gcc options used.
Of note, libsqlite-sys.so.2.8.15 is not reported by wsdiff.

The initial scope of this change was intended to be all of illumos-gate, but during testing it was found that this flagged up warnings in the userland parts due to pulling in out-of-gate components such as perl and openssl.

In the end, the warnings were only enabled for uts/ and changes were limited to the same area plus a few small parts of userland that were highlighted by gcc10's checks.