freebsd stable - Sep 2003 - ports/55928: vmware2 broken on -STABLE, presumably by PAE import

Hi,

On Sun, Aug 24, 2003 at 07:14:29AM -0400, Brandon S. Allbery KF8NH
wrote:
> >Synopsis:       vmware2 broken on -STABLE, presumably by PAE import
>> (..)
Your patch makes VmWare2 build, but it doesn't make it work:

- VmWare fails to allocate memory (see
http://anders.fix.no/test/vmware/1.png and
http://anders.fix.no/test/vmware/2.png).
- There seems to be something wrong with vmmon, some error messages are
reported to my console log:

Console log for noname.aftenposten.no
/dev/vmmon: HostIF_LockPage vpn=0x048a65 mpn=000000 already tracked
Sep  8 10:07:54 noname /kernel: /dev/vmmon: HostIF_LockPage vpn=0x048a65
mpn=000000 already tracked
/dev/vmmon: HostIF_LockPage vpn=0x048a65 mpn=000000 already tracked
/dev/vmmon: HostIF_LockPage vpn=0x048a65 mpn=000000 already tracked
/dev/vmmon: HostIF_LockPage vpn=0x048a65 mpn=000000 already tracked
/dev/vmmon: HostIF_LockPage vpn=0x048a65 mpn=000000 already tracked
Sep  8 10:07:58 noname last message repeated 4 times

Oh, and FreeBSD gets unstable with these patches of yours. I've seen
several hard hangs.

I cleaned up the patch to be relative to the current port (patch
attached). I'm applying your patch (pmap.patch-stable) if OSVERSION is
less than 500000, and higher or equal to 480101 (which was used at the
time of the PAE MFC).

For those on the Cc: list, I added you because you committed PAE/pmap
things, are on the -stable list or maintain the vmware2 port. :)

Any tips or improvements to the patch to make VmWare2 work in FreeBSD
again is very welcome. If we want VmWare supported in -stable at all
(4.9 being right around the corner and all), this must be dealt with
ASAP.

Feel free to use me for testing your suggestions.

Cheers,

-- 
Anders.
-------------- next part --------------
diff -Nur vmware2.old/Makefile vmware2/Makefile
--- vmware2.old/Makefile	Sun Apr 13 11:55:43 2003
+++ vmware2/Makefile	Mon Sep  8 09:22:11 2003
@@ -99,6 +99,8 @@
 .endif
 .if ${OSVERSION} >= 500027
 	${CAT} ${FILESDIR}/pmap.patch | (cd ${WRKSRC} && patch) > /dev/null
2>&1
+.elseif ${OSVERSION} < 500000 && ${OSVERSION} >= 480101
+	${CAT} ${FILESDIR}/pmap.patch-stable | (cd ${WRKSRC} && patch) >
/dev/null 2>&1
 .endif
 	${CP} ${FILESDIR}/Makefile ${WRKSRC}
 	${CP} ${FILESDIR}/Makefile.vmmon ${WRKSRC}/vmmon-only/Makefile
diff -Nur vmware2.old/files/pmap.patch-stable vmware2/files/pmap.patch-stable
--- vmware2.old/files/pmap.patch-stable	Thu Jan  1 01:00:00 1970
+++ vmware2/files/pmap.patch-stable	Mon Sep  8 09:17:55 2003
@@ -0,0 +1,16 @@
+--- vmmon-only/freebsd/hostif.c.orig	Mon Sep  8 09:10:39 2003
++++ vmmon-only/freebsd/hostif.c	Mon Sep  8 09:11:28 2003
+@@ -181,11 +181,11 @@
+ {
+ #define DEB(x) 
+    caddr_t addr = (caddr_t)VPN_2_VA(ppn);
+-   pt_entry_t pteptr = (pt_entry_t)vtopte(addr);
++   pt_entry_t pteptr = (pt_entry_t)vtopte((unsigned long) addr);
+    PTE pte;
+ 
+    DEB(printf("FindMPN: for page %d address %p(phys %p) pteptr %p",
ppn, addr, (caddr_t)vtophys(addr), pteptr));
+-   pte=*pteptr;
++   pte=*((caddr_t)pteptr);
+    DEB(printf("(0x%08x)\n", pte));
+    if (pte & PTE_P) { 
+       return PTE_2_PFN(pte);

On Mon, 2003-09-08 at 05:21, Anders Nordby wrote:
> Your patch makes VmWare2 build, but it doesn't make it work:
I mentioned that in the bug report (and also said "almost certainly
wrong"...).
> Oh, and FreeBSD gets unstable with these patches of yours. I've seen
> several hard hangs.
I wouldn't have noticed as my FreeBSD was unstable anyway; this was
before any patches for the PAE code appeared, so my machine was crashing
every few hours anyway.  :/

In any case, I had already guessed from the type change away from
pointers that just using the provided values would do the wrong thing. 
I was hoping that getting the module to compile would at least produce
some useful diagnostics before the incorrect page manipulations trashed
too much of kernel memory....  (I am no kernel hacker, and in particular
know approximately nothing about FreeBSD's memory management or how the
PAE import changed it.)

-- 
brandon s. allbery    [linux,solaris,freebsd,perl]     allbery@kf8nh.com
system administrator      [WAY too many hats]        allbery@ece.cmu.edu
electrical and computer engineering, carnegie mellon univ.         KF8NH
URGENT!  E-xpedient nuked APK subdomains; kf8nh.apk.net is DEAD.  Sorry.