Date: Thu, 25 May 2023 07:03:14 -0700 From: John Baldwin <jhb@FreeBSD.org> To: Hans Petter Selasky <hps@selasky.org>, Mark Millard <marklmi@yahoo.com> Cc: Jessica Clarke <jrtc27@freebsd.org>, freebsd-arch <freebsd-arch@freebsd.org> Subject: Re: [RFC] An idea for general kernel post-processing automation in FreeBSD Message-ID: <46e55cd8-a781-32d9-17b4-d836949c58e1@FreeBSD.org> In-Reply-To: <da887293-5e29-e06e-bed5-b55acf73220c@selasky.org> References: <2EDDC5DC-81C2-4EB8-B729-66F03A8854E4.ref@yahoo.com> <2EDDC5DC-81C2-4EB8-B729-66F03A8854E4@yahoo.com> <6293f06b-927f-432a-3911-808b1d99441b@selasky.org> <B8217B6F-DA83-4F90-B87B-37E3397237B2@yahoo.com> <cc078296-76b0-9a27-e037-8a25d1e2e9cd@selasky.org> <C7699C53-4961-4C57-82D0-91B468EAA302@yahoo.com> <9C0CE0A5-150D-4FE1-A838-F1E6A39960F6@yahoo.com> <204FCA67-3FCD-48BA-A373-ABE8AD915D40@yahoo.com> <dc56253b-885c-886d-6658-dd3cf204aac4@selasky.org> <738F6620-E4FA-4960-87D2-61B93921593C@yahoo.com> <614513c9-06c0-0330-2969-ad4f3ca06569@selasky.org> <da887293-5e29-e06e-bed5-b55acf73220c@selasky.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 5/25/23 4:49 AM, Hans Petter Selasky wrote: > On 5/25/23 13:42, Hans Petter Selasky wrote: >> On 5/25/23 12:57, Mark Millard wrote: >>> The pre-existing code expresses explicitly that no other >>> routine is allowed to have its own use of the mutex, a >>> design choice enforced by the compiler as things are >>> written. (The purpose of the limitation to block scope.) >>> >>> Your proposed change removes the compiler enforcement of >>> that design, allowing use of the mutex by other code in >>> mlx4_main.c without any notification by the compiler. >>> >>> Your proposal has a direction of being more fragile for >>> bad changes without having to be explicit in code updates >>> about the change of status. >> >> Hi Mark, >> >> Looking only at the mutex part alone, you are right, but not when also >> considering the SYSINIT() part, as implemented in LinuxKPI currently. > > To be more precise: > > The static mutex can only be accessed from within the routine itself, > when it is part of a block scope. That is expected. > > However the static sysinit, which is also inside the block scope of the > function, is accessed from _outside_ the function. > This might be viewed as a violation of the block scope limitation. > > Therefore the DEFINE_MUTEX() should be outside the block scope, due to > how it is implemented in the LinuxKPI currently. I don't think you are correct here. A sysinit is much like a constructor, and it's perfectly valid to pass a pointer to an object that can not be accessed by name outside of a scope to some other bit of code outside of that scope (e.g. a constructor). It's not even just the SYSINIT. You are passing a pointer to this function-static variable to routines like sx_xlock, etc. as well which is also outside of scope, but that's perfectly fine. The scoping rules are about when the compiler can resolve the variable by name, not restricting which bits of code are able to access the variable by reference/pointer. However, I also think that this sub-thread has passed its point of usefulness and should end. -- John Baldwin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?46e55cd8-a781-32d9-17b4-d836949c58e1>