Date: Thu, 1 Feb 2001 09:13:25 -0800 From: "Sam Leffler" <sam@errno.com> To: "Andrew Gallatin" <gallatin@cs.duke.edu>, <freebsd-emulation@FreeBSD.ORG> Cc: <becker@cs.duke.edu> Subject: Re: experiences running vmware w/freebsd 5.0-current as guest Message-ID: <06d701c08c72$49a9d080$24a6d4d1@melange> References: <14968.51758.59870.15884@grasshopper.cs.duke.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
----- Original Message -----
From: "Andrew Gallatin" <gallatin@cs.duke.edu>
To: <freebsd-emulation@FreeBSD.ORG>
Cc: <becker@cs.duke.edu>
Sent: Wednesday, January 31, 2001 6:43 PM
Subject: experiences running vmware w/freebsd 5.0-current as guest
>
> Just a heads up so this makes it into the archives:
>
> When running FreeBSD-current as a guest in vmware (VMX86:
> version='unreleased' build='$Name: build-570 $' option=Release2.0.1) I
> ran into 2 problems:
>
> a) VMware crashes when the pcn driver probes the AMD PCInet card.
> This is indicated by a "NOT_IMPLEMENTED F(562):1654" message
> shortly after FreeBSD probes the PCI bus, right after
> message pci0: <display> at 15.0 (no driver attached).
>
This assertion is caused by the pcn driver doing 32-bit ins and outs to
16-bit registers (the upper 16 bits are reserved).
> Switching to a kernel without the pcn driver driver works around this.
> The lnc driver mostly still works, but is unable to correctly
> determine the card's mac address. This doesn't seem to matter for
> running host-only networking.
>
> b) 5.0-current is _slow_. Mounting root takes upwards of 30 seconds
> on a 400MHz mobile PII. Going multi-user takes over 10 minutes.
> This is compared to near-native perf. with a 4.2-RELEASE kernel.
>
> Taking the i386 version of the atomic_cmpset_int() inline in
> sys/i386/include/atomic.h is an effective workaround for this. Note
> that it is apparently not sufficient to build an I386_CPU kernel - I
> had trouble booting this. Taking just the i386 version of
> atomic_cmpset_int() seems to work just fine, though.
>
> My theory is that the cmpxchgl instruction causes some sort of trap to
> software emulation, but I have not verified it.
>
It could be that the use of the cmpxchgl instruction is finding a slow path
through the runtime; if you have the code fragment handy I can forward it to
the appropriate VMware folks.
Sam
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-emulation" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?06d701c08c72$49a9d080$24a6d4d1>