gcc9: quiet Waddress-of-packed-member for kernel build

[rlibby]

Summary

This is lame, but it's what we already do for the clang build. We take
misaligned pointers into network header structures in many places.

Test Plan

make CROSS_TOOLCHAIN=amd64-gcc9 buildkernel

[• ian]

Yeah, it kinda looks like gcc 9 is just catching up by adding a warning clang has had for a while.

[jhb]

Maybe include share/mk/bsd.sys.mk at the same time as userland needs it as well (at least it did for me)?

[rlibby]

In D22876#501221, @jhb wrote:

Maybe include share/mk/bsd.sys.mk at the same time as userland needs it as well (at least it did for me)?

Yes, I just ended up putting it in a different review, D22889, because I wasn't sure at first how widespread it would be.

[imp]

head/sys/conf/kern.mk

75

This should have been no-error... we want the warnings, just don't want them to kill the build.

[rlibby]

I can prepare a follow up. What I did here was exactly what we already do for clang. I will change both. And then I assume you'd have the same comment for the user build?

[imp]

In D22876#501313, @rlibby wrote:

I can prepare a follow up. What I did here was exactly what we already do for clang. I will change both. And then I assume you'd have the same comment for the user build?

Before we do that, I guess I should ask how noisy that makes the build? If the clang warnings broke with usual practice then there might have been a reason.

[jhb]

I had followed what we did for clang as well in my local patches. I think it is an overly noisy warning. It basically warns if you ever take a pointer to any member that isn't assigned to a void * or a char * pointer (since it wants an alignment of 1). Often times we use packed for structures to fix a layout (e.g. structures mandated by hardware), but the structures themselves are aligned and the members have suitable alignment within the structure such that the warning is a false positive.

[imp]

In D22876#501377, @jhb wrote:

I had followed what we did for clang as well in my local patches. I think it is an overly noisy warning. It basically warns if you ever take a pointer to any member that isn't assigned to a void * or a char * pointer (since it wants an alignment of 1). Often times we use packed for structures to fix a layout (e.g. structures mandated by hardware), but the structures themselves are aligned and the members have suitable alignment within the structure such that the warning is a false positive.

Sometimes I think we should tag the expected alignment. For x86 it doesn't matter, but for arm64 there may be a few that point out real bugs... we do need to be careful, though, because we don't want to pessimize accesses by getting the alignment wrong.

So that sounds like a good reason to depart for now....

[rlibby]

Here's what changing it to no-error for the clang build does (amd64):

buildkernel KERNCONF=GENERIC:

vali% zgrep address-of-packed-member buildkernel-noerror.1.log.gz | wc -l                                               
     524
vali% zgrep address-of-packed-member buildkernel-noerror.1.log.gz | sed 's/.*: warning: //' | sort | uniq -c | sort -n | tail -n 5
  16 taking address of packed member 'ip_dst' of class or structure 'ip' may result in an unaligned pointer value [-Waddress-of-packed-member]
  17 taking address of packed member 'mld_addr' of class or structure 'mld_hdr' may result in an unaligned pointer value [-Waddress-of-packed-member]
  18 taking address of packed member 'ip_src' of class or structure 'ip' may result in an unaligned pointer value [-Waddress-of-packed-member]
 185 taking address of packed member 'ip6_dst' of class or structure 'ip6_hdr' may result in an unaligned pointer value [-Waddress-of-packed-member]
 186 taking address of packed member 'ip6_src' of class or structure 'ip6_hdr' may result in an unaligned pointer value [-Waddress-of-packed-member]

buildworld:

vali% zgrep address-of-packed-member buildworld-noerror.1.log.gz | wc -l                                                         
      59
vali% zgrep address-of-packed-member buildworld-noerror.1.log.gz | sed 's/.*: warning: //' | sort | uniq -c | sort -n 
   1 taking address of packed member 'rpu_prefix' of class or structure 'rr_pco_use' may result in an unaligned pointer value [-Waddress-of-packed-member]
   2 taking address of packed member 'rpm_prefix' of class or structure 'rr_pco_match' may result in an unaligned pointer value [-Waddress-of-packed-member]
   3 taking address of packed member 'nd_opt_pi_prefix' of class or structure 'nd_opt_prefix_info' may result in an unaligned pointer value [-Waddress-of-packed-member]
   3 taking address of packed member 'tss_ist1' of class or structure 'amd64tss' may result in an unaligned pointer value [-Waddress-of-packed-member]
   4 taking address of packed member 'ip6_src' of class or structure 'ip6_hdr' may result in an unaligned pointer value [-Waddress-of-packed-member]
   4 taking address of packed member 'ip_dst' of class or structure 'ip' may result in an unaligned pointer value [-Waddress-of-packed-member]
   5 taking address of packed member 'ip6_dst' of class or structure 'ip6_hdr' may result in an unaligned pointer value [-Waddress-of-packed-member]
  10 taking address of packed member 'ip_src' of class or structure 'ip' may result in an unaligned pointer value [-Waddress-of-packed-member]
  27 taking address of packed member 'status' of class or structure 'nvme_completion' may result in an unaligned pointer value [-Waddress-of-packed-member]

The kernel build would be pretty noisy. In the kernel, there are many places where we are reaching into an mbuf, and I think they really may not be aligned. The user build is maybe not "too" noisy.