VirtualBox on Solaris 11.4 systems with Intel Haswell or later causes host panic due to an SMAP violation => fixed in svn/6.1.x x>18

[dstes]

Hello,

is Solaris 11.4 beta supported please as host OS ?

VirtualBox-5.2.18-SunOS-amd64-r124319.pkg

cat /etc/release 
                             Oracle Solaris 11.4 X86
  Copyright (c) 1983, 2018, Oracle and/or its affiliates.  All rights reserved.
                             Assembled 02 July 2018

pkg list entire
NAME (PUBLISHER)                                  VERSION                    IFO
entire                                            11.4-11.4.0.0.1.10.0       i--

When installing on Solaris 11.4 beta I get a FAILED message :

Detected Solaris 11 Version 175
Loading VirtualBox kernel modules...
   - Added: Host driver
   - Added: NetAdapter driver
   - Added: NetFilter (Crossbow) driver
   - Added: USBMonitor driver
   - Added: USB driver
Configuring services...
svcadm: svc:/application/virtualbox/zoneaccess:default is misconfigured ("restar
ter" property group lacks "state" property).
   * Warning!! Enabling Zone access service  ...FAILED.
   * Warning!! Refer /var/svc/log/application-virtualbox-zoneaccess:default.log 
for details.
Installing MIME types and icons...
Installing Python bindings...
   - Installed: Bindings for Python 2.7
Updating the boot archive...

Installation of <SUNWvbox> was successful.

Unfortunately while trying to create a VM guest , the host panics:

vmcore.0> ::stack
0xfffffffff835a391()
0xfffffffff8354ae5()
0xfffffffff8355e6a()
0xfffffffff83fefa6()
0xfffffffff8356a30()
supdrvIOCtl+0x156f()
VBoxDrvSolarisIOCtl+0x375()
cdev_ioctl+0x67(10f00000003, c0185687, 7fffbcffec50, 202003, ffffa1c0215441a0, 
ffffe330027dde14)
spec_ioctl+0x62(ffffa1c020d90880, c0185687, 7fffbcffec50, 202003, 
ffffa1c0215441a0, ffffe330027dde14)
fop_ioctl+0xf1(ffffa1c020d90880, c0185687, 7fffbcffec50, 202003, 
ffffa1c0215441a0, ffffe330027dde14)
ioctl+0x18a(f, ffffffffc0185687, 7fffbcffec50)
sys_syscall+0x247()

Is this a known issue please ? I could try a test build ...

Thank you David

[dstes]

VirtualBox-5.2.97-SunOS-amd64-r124236.pkg installs without error but the host is still panic'ing when lauching a guest vm.

[dstes]

I'm trying to recompile the latest SVN source on Oracle 11.4. There is a GCC 7.3. There is no libIDL but fortunately this compiles easily on solaris 11.4. Checking for libIDL: found version 0.8.14, OK. It now seems I need QT5, I compiled Qt-4.8.7 that went relatively well, I'll give QT5 a try although that I thought based on the Solaris build instructions that QT4 was needed.

[dstes]

I'll try first to manually set QTDIR to the 4.8.7; using ./configure --disable-hardening --disable-sdl --disable-qt configures

when building it now fails on

# include <sys/old_procfs.h>

in src/VBox/Runtime/r3/solaris/coredumper-solaris.cpp

[dstes]

it seems it is not possible to avoid old_procfs due to

prpsinfo_t ProcInfoOld; /< Process info. Older version (for GDB compat.) */

and because old_procfs.h is gone in sol11.4 that seems to be a show stopper

[Socratis]

Related discussion in the forums: ​https://forums.virtualbox.org/viewtopic.php?f=11&t=89293

[dminer]

This is caused by VirtualBox running into the Solaris SMAP. If you add -B smap-enable=0 to the host's kernel boot arguments you should be OK to launch guests.

[dstes]

Thanks for the hint - indeed there is a note in the Solaris 11.4 release notes : Panic via supdrvioctl and vboxdrvsolarisioctl (24812986)

Running VirtualBox on an Oracle Solaris x64 host system with Supervisor Mode Access Prevention (SMAP) enabled might panic the host with a message similar to the following:

BAD TRAP: type=e (#pf Page fault) rp=fffffffc802c98e0 addr=ffff80ffbc8ff5e0 occurred in module "<unknown>" due to an illegal access to a user address

Workaround: Run the sxadm disable smap command and reboot before starting VirtualBox.

[dstes]

Much better now, no crash / panic any longer on Solaris 11.4 after setting smap to 'disabled' :

# sxadm status EXTENSION STATUS FLAGS aslr enabled (tagged-files) u-c-- nxstack enabled (all) u-c-- nxheap enabled (tagged-files) u-c-- kpti enabled -kcr- ibpb enabled -kcr- ibrs enabled -kcr- smap disabled -kcr-

I tested this on Solaris 11.4 release:

# pkg list entire NAME (PUBLISHER) VERSION IFO entire 11.4-11.4.0.0.1.15.0 i--

on the other hand I think I have an issue with the vboxnet0 network but will try to test with a "etherstub" network other than vboxnet0

dladm show-phys LINK MEDIA STATE SPEED DUPLEX DEVICE net0 Ethernet up 1000 full bge0 net1 Ethernet unknown 0 unknown bge1 vboxnet0 Ethernet up 1000 full vboxnet0

[dstes]

The solution seems to work fine and I'm very impressed both by Solaris 11.4 and the "kernel zone" technology and with VirtualBox. The network solution that I have used so far is VirtualBox "internal network" (intnet) which seems to work and also "Bridged adapter" which seems to allow me to select an etherstub "est0" adapter that I created as follows : dladm create-etherstub est0 . On the host side on top of est0 I can dladm create-vnic -l est0 vnic0 and assign an ip to it : ipadm create-ip vnic0 and ipadm create-addr -a 192.168.0.1/24 vnic0/v4. Then in a VM with the "bridged adapter" set to est0 I can assign an ip address like 192.168.0.2 to the VM and this seems to work, I can "ssh" from the host into the vm and vice versa, the traffic seems to go over the etherstub. In any case, the VirtualBox 5.2.20 seems to work on Solaris 11.4.

[Michal Nowak]

We see this OS crash on OpenIndiana frequently with SMAP-aware CPUs. I noticed that several such SMAP issues were fixed VirtualBox for Linux and macOS for 4.x and 5.x. Is there a plan to fix it for Solaris/OpenSolaris too?

Disabling SMAP in the OS works.

Linux and macOS fixes:

https://www.virtualbox.org/ticket/14412 https://www.virtualbox.org/ticket/13961 https://www.virtualbox.org/ticket/13951 https://www.virtualbox.org/ticket/13820

panic[cpu2]/thread=ffffff09fcbcdbc0: 
BAD TRAP: type=e (#pf Page fault) rp=ffffff002f7a38e0 addr=fffffd7fe934d8e0 occurred in module "<unknown>" due to an illegal access to a user address


VirtualBoxVM: 
#pf Page fault
Bad kernel fault at addr=0xfffffd7fe934d8e0
pid=1371, pc=0xfffffffff818ee06, sp=0xffffff002f7a39d8, eflags=0x10282
cr0: 80050033<pg,wp,ne,et,mp,pe>  cr4: 3606f8<smap,smep,osxsav,pcide,xmme,fxsr,pge,mce,pae,pse,de>
cr2: fffffd7fe934d8e0  
cr3: 6417a6000  
cr8: 0

        rdi:               1a rsi:                1 rdx:                0
        rcx: ffffff002f7a39ef  r8:                0  r9: fffffd7fe934d5a0
        rax: fffffd7fe934d8e0 rbx: ffffff0a11217000 rbp: ffffff002f7a3a00
        r10:               34 r11:               34 r12: ffffff0a0d275dd0
        r13:         beef0000 r14:                0 r15:                0
        fsb: fffffd7fef138a40 gsb: ffffff09f3437000  ds:               4b
         es:               4b  fs:                0  gs:                0
        trp:                e err:                1 rip: fffffffff818ee06
         cs:               30 rfl:            10282 rsp: ffffff002f7a39d8
         ss:               38

ffffff002f7a37e0 unix:die+c6 ()
ffffff002f7a38d0 unix:trap+11b1 ()
ffffff002f7a38e0 unix:cmntrap+e6 ()
ffffff002f7a3a00 fffffffff818ee06 ()
ffffff002f7a3a70 fffffffff818a500 ()
ffffff002f7a3a90 fffffffff818ae45 ()
ffffff002f7a3b30 fffffffff8249d06 ()
ffffff002f7a3bb0 fffffffff818ae15 ()
ffffff002f7a3c40 vboxdrv:supdrvIOCtl+154f ()
ffffff002f7a3cd0 vboxdrv:VBoxDrvSolarisIOCtl+361 ()
ffffff002f7a3d10 genunix:cdev_ioctl+2b ()
ffffff002f7a3d60 specfs:spec_ioctl+45 ()
ffffff002f7a3df0 genunix:fop_ioctl+5b ()
ffffff002f7a3f00 genunix:ioctl+143 ()
ffffff002f7a3f10 unix:brand_sys_syscall+1fe ()

More info to be attached.

[Michal Nowak]

OpenIndiana (illumos) crash info

[Frank Batschulat (Oracle)]

​https://docs.oracle.com/cd/E37838_01/html/E60973/appcompat.html#scrolltoc

Panic via supdrvioctl and vboxdrvsolarisioctl (24812986)

Running VirtualBox on an Oracle Solaris x64 host system with Supervisor Mode Access Prevention (SMAP) enabled might panic the host with a message similar to the following:

BAD TRAP: type=e (#pf Page fault) rp=fffffffc802c98e0 addr=ffff80ffbc8ff5e0 occurred in module "<unknown>" due to an illegal access to a user address

Workaround: Run the sxadm disable smap command and reboot before starting VirtualBox.

[Michal Nowak]

@fbatschu: Thanks, I am aware of the workaround, I was more interested in fix in VirtualBox. (I suspect problem in VirtualBox because the problem is the same on Solaris and illumos despite different SMAP implementations.)

[paulson]

This has been addressed in revision r143291 of the development branch of VirtualBox and is available now in the development snapshot Testbuilds. The fix has also been backported to the 6.1 branch of VirtualBox as of revision r143350 and is also available now in the 6.1.x Testbuilds. The fix will also be part of the next maintenance release of VirtualBox 6.1.x.

Feel free to try them out and let us know if any issues are encountered with the updated builds.