Re: kern/53666: tcpdump for i386 does not work with COMPAT_NETBSD32 on amd64

[Rin Okuyama <rokuyama%rk.phys.keio.ac.jp@localhost>]

The following reply was made to PR kern/53666; it has been noted by GNATS.

From: Rin Okuyama <rokuyama%rk.phys.keio.ac.jp@localhost>
To: Paul Goyette <paul%whooppee.com@localhost>
Cc: matthew green <mrg%eterna.com.au@localhost>, gnats-bugs%NetBSD.org@localhost,
 kern-bug-people%netbsd.org@localhost, gnats-admin%netbsd.org@localhost, netbsd-bugs%netbsd.org@localhost
Subject: Re: kern/53666: tcpdump for i386 does not work with COMPAT_NETBSD32
 on amd64
Date: Mon, 15 Oct 2018 18:55:55 +0900

 Sorry for the late reply,
 
 On 2018/10/12 18:09, Paul Goyette wrote:
 > On Fri, 12 Oct 2018, Rin Okuyama wrote:
 ...
 >> Is it OK to have #ifdef COMPAT_50 (or equivalently,
 >> #ifdef BIOC[GS]ORTIMEOUT) in netbsd32_ioctl.c?
 > 
 > Oh, joy!  Here comes another conflict with the pgoyette-compat
 > branch!
 > 
 > But YES, for now it would be acceptable to have a #ifdef COMPAT_50
 > block.  I will extract the conditional code and move it into
 > compat/netbsd32/netbsd32_compat_50.c and provide an appropriate
 > linkage/hook.
 > 
 > (Actually, this prompted me to look at the existing code on my
 > branch, and it seems I've still got some clean-up work to do here
 > for compat32_50_rnd_ioctl() - the "hook" needs to use the new
 > MP-safe mechanism to prevent the module code from being unloaded
 > while the code is executing!)
 
 and sorry for troubling you...
 
 >> Anyway, I would like to decide the way before proceeding further.
 >>
 >> (1) add unused "u_long cmd" argument to all netbsd32_{to,from}_foo()
 >> functions in netbsd32_conv.h, and move netbsd32_{to,from}_foo() in
 >> netbsd32_ioctl.c to netbsd32_conv.h
 >>
 >> or
 >>
 >> (2) leave preprocessor mess as is
 >>
 >> or
 >>
 >> (3) other
 >>
 >> Paul, how do you think?
 > 
 > Neither 1 nor 2 "feels right" to me.  But I don't have any better
 > choice to offer.
 
 Hmm, unfortunately, support for BIOC[GS]ORTIMEOUT makes the situation
 worse. We have netbsd32_{to,from}_timeval50() in netbsd32_conv.h, and
 they convert between netbsd32_timeval50 and timeval, not timeval50!
 Apparently, we have too many different kinds of functions that share
 the same naming rule :(.
 
 Fix may be made in two steps: (1) minimum fix to HEAD for now and
 netbsd-8, in order to reduce conflicts b/w pgoyette-compat branch as
 much as possible, then (2) real fix after merge of pgoyette-compat.
 
 However, I'm not sure neither whether "the minimum fix" like this
 
    http://www.netbsd.org/~rin/netbsd32_ioctl_20181015.patch
 
 is acceptable, nor how "the real fix" should be...
 
 Thanks,
 rin