Date: Sat, 26 Mar 2011 21:09:43 -0700 (PDT) From: Matthew Dillon <dillon@apollo.backplane.com> To: freebsd-arch@freebsd.org Subject: Re: kernel memory checks on boot vs. boot time Message-ID: <201103270409.p2R49h9e081051@apollo.backplane.com> References: <20110323200200.GA85810@freebsd.org> <201103232050.p2NKov4g017463@lurza.secnetix.de> <4D8A7976.5090103@freebsd.org> <20110323232209.GA15486@freebsd.org> <20110324083048.60862a0f@ernst.jennejohn.org> <20110324214320.K1149@besplex.bde.org>
next in thread | previous in thread | raw e-mail | index | archive | help
In DFly I just changed to code to check one word per 64KB instead of
one word per 4KB, which cuts the time down by a factor of 16. At least
some basic 'is this actually memory' test should probably still be done.
Probing 64G of ram was really annoying before I made that change.
Another solution would be to use a much larger mmu mapping, like a DMAP
style mapping. Iterating the dram isn't going to be fast no matter what
since each test requires a full RAS cycle anyway. I didn't try this
approach but it would be easier to code (treating the page directory
entry as a mapping cache for multiple tests).
Adding programmable bells and whistles would be overkill, and high-end
machines shouldn't require a tunable bypass to boot reasonably anyhow.
-Matt
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201103270409.p2R49h9e081051>