Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 13 Sep 2017 11:24:26 -0600
From:      Warner Losh <imp@bsdimp.com>
To:        Alan Somers <asomers@freebsd.org>
Cc:        "Michael W. Lucas" <mwlucas@michaelwlucas.com>,  "freebsd-hackers@freebsd.org" <hackers@freebsd.org>
Subject:   Re: required kernel rebuilds
Message-ID:  <CANCZdfruXxGvYuk_Vg11maCgTOLdeGZbtyxwAXxKQB_KsC4NbA@mail.gmail.com>
In-Reply-To: <CAOtMX2g6F9nb8xt-h=zWZHRsO9bFvyFaG-DXn9ki8NE49eQQhQ@mail.gmail.com>
References:  <20170913155542.GA25871@mail.michaelwlucas.com> <CAOtMX2g6F9nb8xt-h=zWZHRsO9bFvyFaG-DXn9ki8NE49eQQhQ@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, Sep 13, 2017 at 10:07 AM, Alan Somers <asomers@freebsd.org> wrote:

> On Wed, Sep 13, 2017 at 9:55 AM, Michael W. Lucas
> <mwlucas@michaelwlucas.com> wrote:
> > Hi,
> >
> > Book research question.
> >
> > Way back in the Dark Ages of the 1990s, it wasn't uncommon to rebuild
> > a kernel to fix a recurring panic. You'd have to tune MAXUSERS, or
> > PMAP_SHPGPERPROC.
> >
> > AFAIK, dang near everything is tunable in either loader.conf or
> > sysctl.conf. I really want to say that kernel rebuilds for these kinds
> > of limits aren't needed any more.
> >
> > Does anyone have a counter-example, though? Is anything possibly
> > crash-inducing non-tunable without a kernel rebuild?
> >
> > Thanks,
> > ==ml
>
> Enabling VIMAGE still requires a kernel rebuild, because of the
> possibility of crashes.  I've seen such crashes on 10.2 or 10.3.  I
> don't know if they're fixed in 11.1, but I haven't seen any yet.
> Changing MAXPHYS also requires a kernel rebuild.  AFAIK MAXPHYS isn't
> related to any kernel crashes, but it does have performance
> implications.
>

MAXPHYS isn't currently dynamic. We have to set it at Netflix because we
want to do super large I/Os. We'd just bump it in base, but it requires
extra memory for every single I/O (since there's arrays in struct buf / bio
based on this size for pages related to the I/O).

I've had one kernel panic that was fixed by changing MAXPHYS, but that was
due to a stupid driver bug...

Warner



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CANCZdfruXxGvYuk_Vg11maCgTOLdeGZbtyxwAXxKQB_KsC4NbA>