Date: Fri, 8 Mar 2002 15:49:37 -0800 (PST) From: Matthew Dillon <dillon@apollo.backplane.com> To: Jake Burkholder <jake@locore.ca> Cc: Robert Watson <rwatson@FreeBSD.ORG>, FreeBSD current users <current@FreeBSD.ORG> Subject: Re: Patch for critical_enter()/critical_exit() & interrupt assem Message-ID: <200203082349.g28Nnbi80049@apollo.backplane.com> References: <200203072143.g27LhaL97112@harmony.village.org> <Pine.NEB.3.96L.1020307171254.23264D-100000@fledge.watson.org> <20020307205844.C12044@locore.ca> <200203080304.g2834n571759@apollo.backplane.com> <20020308024653.B14181@locore.ca> <200203080923.g289N1N74926@apollo.backplane.com> <20020308151743.B17591@locore.ca>
next in thread | previous in thread | raw e-mail | index | archive | help
:The reason is that if they are in MI code they automatically apply to all
:platforms and can't get out sync. When they are modified to handle preemption
:the freebsd kernel will be fully preemptive. Not, it works on i386 and its
:believed to work on alpha and powerpc is not preemptive at all and we don't
If the routines were large this would be an issue, but we are talking
between 1 and 5 lines of MI code verses potentially many more lines of
MD code and I find it virtually impossible for such a small amount of
code to 'get out of sync'.
:even know about ia64 (not to dump on other platforms, this is just example of
:how things go sometimes). I understand that this argument may be sentimental
:and may not hold water in a technial discussion. As such I will not stop you
:from making them MD if you disagree (not to imply that I have the power to do
:so, I don't), I just think that keeping them MI is the right thing to do.
:
:I must admit that having them be MD will allow me to make optimizations for
:sparc64 that I have wanted to. However, I do not think that this is better
:for freebsd as a whole.
:
:>
:> :...
:...
:The point is not wether its easy or not, the point is that this is an
:important feature that may have been forgotten about. I use this on
:a daily basis in sparc64 development and I would be upset if it was
:broken there for any amount of time, not that this patch will affect it.
:I am confident that you will fix any problems that arise, but I would
:rather the next person that tries to do some debugging not be confused
:by something being different, or by it not working and having to wait for
:a fix, even if that amount of time is insignificant. If you do not feel
:that this needs to be looked at before you commit then that is fine, again
:I cannot stop you. I know many other committers who would not feel that
:way about a patch of their own and I think that standard is worth adhereing
:to. I apologize if this sounds like a lecture or if this offends you, it
:is just how I feel.
:
:Jake
I'm not sure what you are refering to here. The fast interrupt
deferral stuff only applies to I386.
In anycase, I certainly haven't forgotten about debugger support
for I386, but I didn't go to great lengths to test whether the
DDB backtrace operates as expected for the fast-interrupt-restart
case either.
-Matt
Matthew Dillon
<dillon@backplane.com>
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200203082349.g28Nnbi80049>