Date: Mon, 11 Oct 1999 19:07:06 +0800 From: Peter Wemm <peter@netplex.com.au> To: Matthew Dillon <dillon@apollo.backplane.com> Cc: Poul-Henning Kamp <phk@critter.freebsd.dk>, "Daniel C. Sobral" <dcs@newsguy.com>, cvs-committers@FreeBSD.ORG, cvs-all@FreeBSD.ORG Subject: Re: cvs commit: src/sys/vm vm_swap.c Message-ID: <19991011110706.CF0A01CC8@overcee.netplex.com.au> In-Reply-To: Your message of "Mon, 11 Oct 1999 18:40:06 %2B0800." <19991011104006.AE4631CC8@overcee.netplex.com.au>
next in thread | previous in thread | raw e-mail | index | archive | help
I wrote:
> Matthew Dillon wrote:
> > :>
> > :> We're not removing the device interface. It provides a reasonable
> > :> abstraction and a nice demark between swap_pager.c and vm_swap.c,
> > :> as well as potential flexibility that could be useful in the future
> > :> Removing it will not save time or much in the way of code space.
> > :
> > :The swap *device* is non-functional and doesn't *do* anything except
> > :provide a subroutine call in a very roundabout way. I think it would be
> > :far better to avoid an indirection via the VOP_* system for no useful gain
> > :and do something like the aooended patch (which works perfectly here BTW,
> > :even under heavy swap load on multiple disks). Further (micro)
> > :optimizations are possible, for example pbgetvp() is used to get a p-buffe
r
> > :that's associated with swapdev_vp and the device. The device isn't used
> > :and presently vn_todev(vp) ends up returning NODEV. swapdev_vp is kinda
> > :orphaned with this change but still works ok. p-buffers are created being
> >
> > Gentlemen. You might as well stop. I've said it before and I
> > will say it again: Nobody is screwing around with the swap
> > subsytem. I spent a lot of time on that subsytem, including weeks
> > of testing. I am essentially the maintainer for it, and I am not
> > going to allow people to mess with it based on a few hacks. Just
> > because the swap subsystem uses a small subset of the device
> > interface doesn't mean you can hack it up to remove it. You are not
> > 'fixing' anything by doing so. You aren't making the system more relia
bl
> e,
> > you aren't making the system run better. You are just making my job
> > harder by forcing me to retest everything. So it isn't going to happen
.
>
> Oh come on now! Your "testing" didn't even find that reading /dev/drum was
> an instant guaranteed system panic. This patch does not change a thing
> except remove dead code that is almost certain to break shortly.
Sorry, this was over the top. Seriously though, this isn't part of phk's
"agenda", I was the one that initially suggested it was unused. The swap
"device" is very badly abused at present. It's not opened or closed or
anything like devices are supposed to be. The sole function it provides is
to obfuscate swap_pager.c's calls to the swstrategy() function. The
buffers and swapdev_vp are bound to the underlying physical device, not the
swap device. You aren't the only one that is capable of testing, I have
tested this *trivial* change and it works perfectly. It doesn't conflict
with (or go near) any of your patches so it can't see how it could conflict
with your work.
I challenge you to find a single thing that it breaks.
Cheers,
-Peter
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe cvs-all" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19991011110706.CF0A01CC8>