VirtualBox

Ticket #11035 (closed defect: fixed)

Opened 19 months ago

Last modified 14 months ago

failed to start kernel module vboxdrv (custom kernel) => Fixed in SVN

Reported by: skywk Owned by:
Priority: critical Component: other
Version: VirtualBox 4.2.0 Keywords: custom kernel module starting failed
Cc: Guest type: all
Host type: Linux

Description (last modified by frank) (diff)

The kernel module vboxdrv failed to start, when modprobe vboxdrv, the dmesg shows these:

[   48.500393] vboxdrv: Found 2 processor cores.
[   48.500524] BUG: unable to handle kernel NULL pointer dereference at 00000600
[   48.502986] IP: [<f8de458a>] VBoxHost_RTR0MemObjFree+0x294/0x294 [vboxdrv]
[   48.503333] *pde = 00000000 
[   48.503333] Oops: 0000 [#1] PREEMPT SMP 
[   48.503333] Modules linked in: vboxdrv(O+) coretemp tp_smapi(O) thinkpad_ec(O) loop firewire_sbp2 fuse snd_hda_codec_analog snd_hda_intel snd_hda_codec snd_hwdep thinkpad_acpi snd_pcm snd_page_alloc nvram snd_seq snd_seq_device snd_timer arc4 psmouse serio_raw i2c_i801 rng_core iwl3945 snd iwl_legacy mac80211 cfg80211 soundcore rfkill battery ac evdev processor microcode i915 i2c_algo_bit drm_kms_helper drm i2c_core sg usbhid hid sd_mod crc_t10dif ide_pci_generic uhci_hcd ahci libahci libata scsi_mod piix ide_core sdhci_pci sdhci mmc_core firewire_ohci firewire_core crc_itu_t ehci_hcd usbcore e1000e usb_common video thermal thermal_sys button [last unloaded: scsi_wait_scan]
[   48.503333] 
[   48.503333] Pid: 2324, comm: modprobe Tainted: G           O 3.2.1-ck1-bfs416 #1 LENOVO 636386U/636386U
[   48.503333] EIP: 0060:[<f8de458a>] EFLAGS: 00010293 CPU: 0
[   48.503333] EIP is at VBoxHost_RTR0MemObjGetPagePhysAddr+0x0/0x67 [vboxdrv]
[   48.503333] EAX: f4cac000 EBX: f8dff158 ECX: 00000000 EDX: 00000002
[   48.503333] ESI: f4cac000 EDI: 00000600 EBP: 00000000 ESP: f5673e98
[   48.503333]  DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
[   48.503333] Process modprobe (pid: 2324, ti=f5672000 task=f4f7b930 task.ti=f5672000)
[   48.503333] Stack:
[   48.503333]  f8ddd112 f4e9c850 00000000 f4e9c850 0000e444 00000020 000000d0 00000000
[   48.503333]  00000010 f5673ee0 00000020 f8de4ef3 00000002 00000002 f4caa200 f4caa224
[   48.503333]  f8d72000 f8de3757 f4edcbc0 f8de378a f8de7362 00000010 f8df951f 00000000
[   48.503333] Call Trace:
[   48.503333]  [<f8ddd112>] ? supdrvInitDevExt+0x103/0x697 [vboxdrv]
[   48.503333]  [<f8de4ef3>] ? rtR0MemAllocEx+0x4a/0x9d [vboxdrv]
[   48.503333]  [<f8d72000>] ? 0xf8d71fff
[   48.503333]  [<f8de3757>] ? rtR0MemAlloc+0x8/0x15 [vboxdrv]
[   48.503333]  [<f8de378a>] ? VBoxHost_RTMemAllocTag+0xb/0x18 [vboxdrv]
[   48.503333]  [<f8de7362>] ? VBoxHost_RTSpinlockCreate+0x1b/0x4f [vboxdrv]
[   48.503333]  [<f8de4ca7>] ? rtR0PowerNotificationInit+0x11/0x15 [vboxdrv]
[   48.503333]  [<f8d72000>] ? 0xf8d71fff
[   48.503333]  [<f8d72050>] ? VBoxDrvLinuxInit+0x50/0xc7 [vboxdrv]
[   48.503333]  [<c1001202>] ? do_one_initcall+0x66/0x10c
[   48.503333]  [<c104ffab>] ? sys_init_module+0x136c/0x153f
[   48.503333]  [<c12bff9f>] ? sysenter_do_call+0x12/0x28
[   48.503333] Code: fe ff ff e9 cf fe ff ff 8b 4a 1c 85 c9 0f 84 47 ff ff ff 8d 34 8d fc ff ff ff 89 4c 24 04 e9 5c ff ff ff 83 c4 08 5b 5e 5f 5d c3 <8b> 0f 8b 47 04 8d 91 00 10 00 00 81 fa ff 1f 00 00 76 45 81 39 
[   48.503333] EIP: [<f8de458a>] VBoxHost_RTR0MemObjGetPagePhysAddr+0x0/0x67 [vboxdrv] SS:ESP 0068:f5673e98
[   48.503333] CR2: 0000000000000600
[   48.608313] ---[ end trace 4eecccc0ba5c7023 ]---

I run a custom compiled kernel:

Linux debian.zzx.x60 3.5.4-bfs-424 #2 SMP Thu Sep 27 09:20:10 CST 2012 i686 GNU/Linux

and ...3.2.1-ck1-bfs416...

both failed.

but when run the kernel supplied by debian, i.e. ...3.2.0-3-686-pae... it works fine.

The VBOX version is 4.2.0, 4.1.18 also tried, and is the same as the former.

Attachments

vbox.kernel.log Download (60.0 KB) - added by skywk 19 months ago.
config-3.6.10 Download (92.9 KB) - added by kimrhh 16 months ago.
kernel 3.6.10 .config
vboxdrv.ko Download (244.2 KB) - added by kimrhh 16 months ago.
kernel module failing to load
vboxdrv.ko.gz Download (459.4 KB) - added by kimrhh 16 months ago.
vboxdrv.ko with debug
memobj-r0drv.i.gz Download (215.7 KB) - added by kimrhh 16 months ago.

Change History

Changed 19 months ago by skywk

comment:1 Changed 16 months ago by frank

Still relevant with VBox 4.2.4? This looks to me like a special Linux kernel, does it have some extra patches applied?

comment:2 Changed 16 months ago by kimrhh

I can confirm this bug on VBox 4.2.4

I am running a "vanilla" kernel directly from git from Linus stable branch with no extra patches, .config attached for 3.6.10

I have tried with earlier kernels as well: 3.5.4 and 3.4.2 since I have had vbox running succesfully with those before, but the error is the same.

For me it does not help to downgrade to vbox 4.1.18 though.

Changed 16 months ago by kimrhh

kernel 3.6.10 .config

comment:3 Changed 16 months ago by kimrhh

Forgot something. I've also tried with latest stable kernel: 3.7.0 but with this the vbox modules no longer builds.

Snippet from my dmesg when loading vboxdrv:

vboxdrv: Found 2 processor cores.
BUG: unable to handle kernel NULL pointer dereference at 00000700
IP: [<f977dac2>] VBoxHost_RTR0MemObjFree+0x294/0x294 [vboxdrv]
*pdpt = 0000000026144001 *pde = 0000000000000000 
Oops: 0000 [#1] PREEMPT SMP 
Modules linked in: vboxdrv(O+)
Pid: 14395, comm: modprobe Tainted: G           O 3.6.10 #14 LENOVO 74585NG/74585NG
EIP: 0060:[<f977dac2>] EFLAGS: 00210293 CPU: 1
EIP is at VBoxHost_RTR0MemObjGetPagePhysAddr+0x0/0x67 [vboxdrv]
EAX: e64b6000 EBX: e64b6000 ECX: 264b6000 EDX: 00000002
ESI: f97963c8 EDI: 00000700 EBP: 00000000 ESP: e8ad9e84
 DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
CR0: 8005003b CR2: 00000700 CR3: 30243000 CR4: 000407f0
DR0: 00000000 DR1: 00000000 DR2: 00000000 DR3: 00000000
DR6: ffff0ff0 DR7: 00000400
Process modprobe (pid: 14395, ti=e8ad8000 task=e9f5e6c0 task.ti=e8ad8000)
Stack:
 f9779067 e9dfbb50 00000000 e9dfbb50 c10674fe e9f5e2e0 00000000 00000000
 c10ccb9b 000000d0 f977e4fb 00000000 00000008 e8ad9edc 00000018 f977e4fb
 00000004 00000000 f9796288 e3300f80 f979f000 f977cc8f d4d28960 f977ccc2
Call Trace:
 [<f9779067>] ? supdrvInitDevExt+0xdd/0x72d [vboxdrv]
 [<c10674fe>] ? ttwu_do_wakeup+0xb/0xa3
 [<c10ccb9b>] ? __kmalloc+0x88/0xaa
 [<f977e4fb>] ? rtR0MemAllocEx+0x69/0xbc [vboxdrv]
 [<f977e4fb>] ? rtR0MemAllocEx+0x69/0xbc [vboxdrv]
 [<f979f000>] ? 0xf979efff
 [<f977cc8f>] ? rtR0MemAlloc+0x8/0x15 [vboxdrv]
 [<f977ccc2>] ? VBoxHost_RTMemAllocTag+0xb/0x18 [vboxdrv]
 [<f9780a1c>] ? VBoxHost_RTSpinlockCreate+0xc/0x30 [vboxdrv]
 [<f979f000>] ? 0xf979efff
 [<f979f050>] ? VBoxDrvLinuxInit+0x50/0xc7 [vboxdrv]
 [<c100114e>] ? do_one_initcall+0x66/0x104
 [<c1083912>] ? sys_init_module+0x11d7/0x13de
 [<c15f5bcc>] ? sysenter_do_call+0x12/0x22
Code: fe ff ff e9 cf fe ff ff 8b 4a 1c 85 c9 0f 84 47 ff ff ff 8d 34 8d fc ff ff ff 89 4c 24 04 e9 5c ff ff ff 83 c4 08 5b 5e 5f 5d c3 <8b> 0f 8b 47 04 8d 91 00 10 00 00 81 fa ff 1f 00 00 76 45 81 39
EIP: [<f977dac2>] VBoxHost_RTR0MemObjGetPagePhysAddr+0x0/0x67 [vboxdrv] SS:ESP 0068:e8ad9e84
CR2: 0000000000000700
---[ end trace af95c3bddf7757dd ]---

comment:4 Changed 16 months ago by frank

Please attach your compiled vboxdrv.ko module. Which Linux distribution are you using?

comment:5 Changed 16 months ago by frank

  • Description modified (diff)
  • Summary changed from failed to start kernel module vboxdrv to failed to start kernel module vboxdrv (custom kernel)

Changed 16 months ago by kimrhh

kernel module failing to load

comment:6 Changed 16 months ago by kimrhh

I'm using Exherbo Linux www.exherbo.org

uname -a Linux exherbo-work-laptop 3.6.10 #14 SMP PREEMPT Mon Dec 17 00:32:07 CET 2012 i686 GNU/Linux

Do you have some documentation on debugging this kind of errors with the kernel modules? I've been trying to figure out how to build the kernel module with debug symbols.

attached the module

comment:7 Changed 16 months ago by kimrhh

GNU gdb (GDB) 7.5
Copyright (C) 2012 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "i686-pc-linux-gnu".
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>...
Reading symbols from /usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv.ko...done.
(gdb) list *(supdrvInitDevExt)+0xdd
0x20c4 is in supdrvInitDevExt (/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/SUPDrv.c:4947).
4942         * Allocate a contiguous set of pages with a default kernel mapping.
4943         */
4944        rc = RTR0MemObjAllocCont(&pDevExt->GipMemObj, RT_UOFFSETOF(SUPGLOBALINFOPAGE, aCPUs[cCpus]), false /*fExecutable*/);
4945        if (RT_FAILURE(rc))
4946        {
4947            OSDBGPRINT(("supdrvGipCreate: failed to allocate the GIP page. rc=%d\n", rc));
4948            return rc;
4949        }
4950        pGip = (PSUPGLOBALINFOPAGE)RTR0MemObjAddress(pDevExt->GipMemObj); AssertPtr(pGip);
4951        HCPhysGip = RTR0MemObjGetPagePhysAddr(pDevExt->GipMemObj, 0); Assert(HCPhysGip != NIL_RTHCPHYS);
(gdb)
Last edited 16 months ago by kimrhh (previous) (diff)

comment:8 Changed 16 months ago by frank

To me it looks like the module you attached does not fit the text in comment 3. Can you recheck?

comment:9 Changed 16 months ago by kimrhh

sorry yes, this was rebuild with debug info and perhaps another compiler (been trying to figure out what change made this happen)

Attaching the module build with debug, and the new dmesg as well as information from GDB again.

dmesg:

vboxdrv: Found 2 processor cores.
BUG: unable to handle kernel NULL pointer dereference at 00000700
IP: [<f8c0a55e>] VBoxHost_RTR0MemObjFree+0x294/0x294 [vboxdrv]
*pdpt = 0000000024679001 *pde = 0000000000000000 
Oops: 0000 [#1] PREEMPT SMP 
Modules linked in: vboxdrv(O+)
Pid: 4826, comm: modprobe Tainted: G           O 3.6.10 #17 LENOVO 74585NG/74585NG
EIP: 0060:[<f8c0a55e>] EFLAGS: 00210293 CPU: 0
EIP is at VBoxHost_RTR0MemObjGetPagePhysAddr+0x0/0x67 [vboxdrv]
EAX: e46a2000 EBX: f8c24c60 ECX: 246a2000 EDX: 00000002
ESI: e46a2000 EDI: 00000700 EBP: 00000000 ESP: e5c43e84
 DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
CR0: 8005003b CR2: 00000700 CR3: 2b2d5000 CR4: 000407f0
DR0: 00000000 DR1: 00000000 DR2: 00000000 DR3: 00000000
DR6: ffff0ff0 DR7: 00000400
Process modprobe (pid: 4826, ti=e5c42000 task=e46bd360 task.ti=e5c42000)
Stack:
 f8c030ea eb02d0d0 00000000 eb02d0d0 001ad0b0 f8c0aedb f8c0aedb 00000000
 00000010 e5c43ecc 00000020 f8c0aedb 00000004 00000002 eb0ed6c0 eb0ed6e4
 f8c61000 f8c0972b ee5ceba0 f8c0975e f8c0d436 00000010 f8c1f013 00000000
Call Trace:
 [<f8c030ea>] ? supdrvInitDevExt+0x103/0x697 [vboxdrv]
 [<f8c0aedb>] ? rtR0MemAllocEx+0x69/0xbc [vboxdrv]
 [<f8c0aedb>] ? rtR0MemAllocEx+0x69/0xbc [vboxdrv]
 [<f8c0aedb>] ? rtR0MemAllocEx+0x69/0xbc [vboxdrv]
 [<f8c61000>] ? 0xf8c60fff
 [<f8c0972b>] ? rtR0MemAlloc+0x8/0x15 [vboxdrv]
 [<f8c0975e>] ? VBoxHost_RTMemAllocTag+0xb/0x18 [vboxdrv]
 [<f8c0d436>] ? VBoxHost_RTSpinlockCreate+0x1b/0x4f [vboxdrv]
 [<f8c0ac7b>] ? rtR0PowerNotificationInit+0x11/0x15 [vboxdrv]
 [<f8c61000>] ? 0xf8c60fff
 [<f8c61050>] ? VBoxDrvLinuxInit+0x50/0xc7 [vboxdrv]
 [<c100114e>] ? do_one_initcall+0x66/0x104
 [<c1085f2d>] ? sys_init_module+0x11c7/0x139d
 [<c15f574c>] ? sysenter_do_call+0x12/0x22
Code: fe ff ff e9 cf fe ff ff 8b 4a 1c 85 c9 0f 84 47 ff ff ff 8d 34 8d fc ff ff ff 89 4c 24 04 e9 5c ff ff ff 83 c4 08 5b 5e 5f 5d c3 <8b> 0f 8b 47 04 8d 91 00 10 00 00 81 fa ff 1f 00 00 76 45 81 39
EIP: [<f8c0a55e>] VBoxHost_RTR0MemObjGetPagePhysAddr+0x0/0x67 [vboxdrv] SS:ESP 0068:e5c43e84
CR2: 0000000000000700
---[ end trace dc295d7efa9a1199 ]---

GDB:

GNU gdb (GDB) 7.5
Copyright (C) 2012 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "i686-pc-linux-gnu".
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>...
Reading symbols from /usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv.ko...done.
(gdb) list *(supdrvInitDevExt)+0x103
0x20ea is in supdrvInitDevExt (/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/SUPDrv.c:4951).
4946        {
4947            OSDBGPRINT(("supdrvGipCreate: failed to allocate the GIP page. rc=%d\n", rc));
4948            return rc;
4949        }
4950        pGip = (PSUPGLOBALINFOPAGE)RTR0MemObjAddress(pDevExt->GipMemObj); AssertPtr(pGip);
4951        HCPhysGip = RTR0MemObjGetPagePhysAddr(pDevExt->GipMemObj, 0); Assert(HCPhysGip != NIL_RTHCPHYS);
4952
4953        /*
4954         * Find a reasonable update interval and initialize the structure.
4955         */
(gdb)

Changed 16 months ago by kimrhh

vboxdrv.ko with debug

comment:10 Changed 16 months ago by kimrhh

It took my a while to try this out, but I suspected it was compiler related, and indeed, switching back to GCC 4.5.4 the module loads just fine.

Other compiler versions used was: GCC 4.6.3 and 4.7.2

comment:11 Changed 16 months ago by frank

Yes, I know that modules compiled with some gcc-4.7 compilers are buggy. I still don't know if this is really a compiler bug (I'm using gcc-4.7 from Debian and don't see any issue) or if this is some imprecise code.

comment:12 Changed 16 months ago by frank

Yes, the calling convention of the function RTR0MemObjGetPagePhysAddr() is wrongly respected by gcc. Please recompile the complete vboxdrv.ko module, then touch the file r0drv/memobj-r0drv.c. Then compile again with passing V=1. To compile the module, go directly to the vboxdrv directory. Copy the command line used to compile memobj-r0drv.c here.

Then use the same command line but replace '-c' by '-E -dD' and the suffix of the file name after '-o' from .o to .i. Attention, when you execute this command line then you first have to go to the Linux source code directory (otherwise the pathes will be incorrect). This is usually /lib/modules/uname -r/build.

Then attach the compressed memobj-r0drv.i file to this ticket.

comment:13 Changed 16 months ago by frank

And please use gcc-4.7 to do this experiment.

comment:14 Changed 16 months ago by kimrhh

just to make sure I did it correctly, make V=1 gave me:

gcc -Wp,-MD,/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/r0drv/.memobj-r0drv.o.d  -nostdinc -isystem /usr/lib/gcc/i686-pc-linux-gnu/4.7.2/include -I/usr/src/linux-stable/arch/x86/include -Iarch/x86/include/generated -Iinclude  -include /usr/src/linux-stable/include/linux/kconfig.h -D__KERNEL__ -Wall -Wundef -Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing -fno-common -Werror-implicit-function-declaration -Wno-format-security -fno-delete-null-pointer-checks -Os -m32 -msoft-float -mregparm=3 -freg-struct-return -fno-pic -mpreferred-stack-boundary=2 -march=i686 -mtune=core2 -mtune=generic -Wa,-mtune=generic32 -ffreestanding -DCONFIG_AS_CFI=1 -DCONFIG_AS_CFI_SIGNAL_FRAME=1 -DCONFIG_AS_CFI_SECTIONS=1 -DCONFIG_AS_AVX=1 -pipe -Wno-sign-compare -fno-asynchronous-unwind-tables -mno-sse -mno-mmx -mno-sse2 -mno-3dnow -mno-avx -Wframe-larger-than=2048 -fno-stack-protector -Wno-unused-but-set-variable -fomit-frame-pointer -g -femit-struct-debug-baseonly -Wdeclaration-after-statement -Wno-pointer-sign -fno-strict-overflow -fconserve-stack -DCC_HAVE_ASM_GOTO -include /usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/include/VBox/SUPDrvMangling.h -I/lib/modules/3.6.10/build/include -I/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/ -I/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/include -I/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/r0drv/linux -I/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/vboxdrv/ -I/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/vboxdrv/include -I/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/vboxdrv/r0drv/linux -D__KERNEL__ -DMODULE -DRT_OS_LINUX -DIN_RING0 -DIN_RT_R0 -DIN_SUP_R0 -DVBOX -DRT_WITH_VBOX -DVBOX_WITH_HARDENING -DCONFIG_VBOXDRV_AS_MISC -DRT_ARCH_X86 -DVBOX_WITH_64_BITS_GUESTS  -DMODULE  -D"KBUILD_STR(s)=#s" -D"KBUILD_BASENAME=KBUILD_STR(memobj_r0drv)"  -D"KBUILD_MODNAME=KBUILD_STR(vboxdrv)" -c -o /usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/r0drv/memobj-r0drv.o /usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/r0drv/memobj-r0drv.c

replaced .o with .i, and -c with -e -dD, file attached

Changed 16 months ago by kimrhh

comment:15 Changed 16 months ago by frank

Thank you! This looks definitely like a gcc bug to me. Have a look at line 39219 of memobj-r0drv.i: Here starts the implementation of VBoxHost_RTR0MemObjGetPagePhysAddr(). This function was declared at line 8278. Both, the declaration and the implementation contain the attribute regparm(0) which means that when calling this function, all parameters must be passed using the stack, no parameters must be passed in registers. regparm(3) would be the opposite (allow up to 3 parameters to be passed in registers). Passing parameters using registers is a performance optimization.

When looking at your vboxdrv.ko file with 'objdump -ld vboxdrv.ko' and search for SUPDrv.c:4951, it is clearly visible that both parameters are passed on the stack:

    20de:       6a 00                   push   $0x0
    20e0:       ff 73 4c                pushl  0x4c(%ebx)
    20e3:       89 c6                   mov    %eax,%esi
    20e5:       e8 fc ff ff ff          call   20e6

But the function implementation which starts at address 0x955e at the same file shows that the compiler expects the parameter MemObj as passed in the register edi. Line 272 of r0drv/memobj-r0drv.c is implemented like this

VBoxHost_RTR0MemObjGetPagePhysAddr():
    955e:       8b 0f                   mov    (%edi),%ecx
    9560:       8b 47 04                mov    0x4(%edi),%eax
    9563:       8d 91 00 10 00 00       lea    0x1000(%ecx),%edx
    9569:       81 fa ff 1f 00 00       cmp    $0x1fff,%edx
    956f:       76 45                   jbe    95b6
    ...
    95b6:       83 c8 ff                or     $0xffffffff,%eax
    95b9:       89 c2                   mov    %eax,%edx
    95bb:       c3                      ret    

This is completely bogus and must be a gcc bug. For 32-bit targets, gcc never passes any parameter in register edi:

  • esi, edi, ebx and ebp are callee-saved registers
  • if the function prototype has the regparm(0) attribute, all parameters are passed on the stack
  • if the function prototype has the regparm(3) attribute, up to three parameters can be passed using the register eax, edx and ecx

comment:16 Changed 15 months ago by kimrhh

heh, that is a bit beyond my debugging skills, will you take this to GCC upstream or should I ? It would not be nice to be stuck on gcc-4.5 for vbox :)

comment:17 Changed 15 months ago by frank

Can you tell me which gcc compiler you used to produce this code (complete output of 'gcc -v' please)? And which Linux distribution are you using?

comment:18 Changed 15 months ago by kimrhh

kimrhh@exherbo-work-laptop ~ $ gcc -v
Using built-in specs.                                                                                                                                                                                                                        
COLLECT_GCC=gcc                                                                                                                                                                                                                              
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/i686-pc-linux-gnu/4.7.2/lto-wrapper                                                                                                                                                                     
Target: i686-pc-linux-gnu                                                                                                                                                                                                                    
Configured with: /var/tmp/paludis/build/sys-devel-gcc-4.7.2-r2/work/gcc-4.7.2/configure --prefix=/usr --host=i686-pc-linux-gnu --build=i686-pc-linux-gnu --mandir=/usr/share/man --infodir=/usr/share/info --datadir=/usr/share --sysconfdir=/etc --localstatedir=/var/lib --disable-dependency-tracking --disable-silent-rules --enable-fast-install --libdir=/usr/lib --disable-multilib --libdir=/usr/lib --enable-clocale=gnu --enable-languages=c++,fortran,java --enable-libstdcxx-pch=yes --enable-nls --program-suffix=-4.7 --enable-lto --with-pkgversion='Exherbo gcc-4.7.2-r2' --with-system-zlib --disable-graphite --without-cloog --without-ppl --enable-libgomp --disable-libssp                                        
Thread model: posix                                                                                                                                                                                                                          
gcc version 4.7.2 (Exherbo gcc-4.7.2-r2)

I'm using Exherbo Linux: www.exherbo.org

comment:19 Changed 15 months ago by frank

Ok. But before I file the bug report, are you sure that there is no update for gcc on your distribution?

comment:20 Changed 15 months ago by kimrhh

yes, this is our newest version, and the latest release from GCC

comment:21 Changed 15 months ago by frank

I've created bug  55940 in the gcc bugzilla.

kimrhh, could you also attach the command line (may be as text file) which you used to create the memobj-r0drv.i file? I would like to know the exact command line.

comment:22 Changed 15 months ago by kimrhh

it is already in comment #14 sort of, if I understand you correctly, but here you go (with replaced arguments)

gcc -Wp,-MD,/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/r0drv/.memobj-r0drv.o.d  -nostdinc -isystem /usr/lib/gcc/i686-pc-linux-gnu/4.7.2/include -I/usr/src/linux-stable/arch/x86/include -Iarch/x86/include/generated -Iinclude  -include /usr/src/linux-stable/include/linux/kconfig.h -D__KERNEL__ -Wall -Wundef -Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing -fno-common -Werror-implicit-function-declaration -Wno-format-security -fno-delete-null-pointer-checks -Os -m32 -msoft-float -mregparm=3 -freg-struct-return -fno-pic -mpreferred-stack-boundary=2 -march=i686 -mtune=core2 -mtune=generic -Wa,-mtune=generic32 -ffreestanding -DCONFIG_AS_CFI=1 -DCONFIG_AS_CFI_SIGNAL_FRAME=1 -DCONFIG_AS_CFI_SECTIONS=1 -DCONFIG_AS_AVX=1 -pipe -Wno-sign-compare -fno-asynchronous-unwind-tables -mno-sse -mno-mmx -mno-sse2 -mno-3dnow -mno-avx -Wframe-larger-than=2048 -fno-stack-protector -Wno-unused-but-set-variable -fomit-frame-pointer -g -femit-struct-debug-baseonly -Wdeclaration-after-statement -Wno-pointer-sign -fno-strict-overflow -fconserve-stack -DCC_HAVE_ASM_GOTO -include /usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/include/VBox/SUPDrvMangling.h -I/lib/modules/3.6.10/build/include -I/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/ -I/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/include -I/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/r0drv/linux -I/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/vboxdrv/ -I/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/vboxdrv/include -I/usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/vboxdrv/r0drv/linux -D__KERNEL__ -DMODULE -DRT_OS_LINUX -DIN_RING0 -DIN_RT_R0 -DIN_SUP_R0 -DVBOX -DRT_WITH_VBOX -DVBOX_WITH_HARDENING -DCONFIG_VBOXDRV_AS_MISC -DRT_ARCH_X86 -DVBOX_WITH_64_BITS_GUESTS  -DMODULE  -D"KBUILD_STR(s)=#s" -D"KBUILD_BASENAME=KBUILD_STR(memobj_r0drv)"  -D"KBUILD_MODNAME=KBUILD_STR(vboxdrv)" -e -dD -o /usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/r0drv/memobj-r0drv.i /usr/src/virtualbox-bin-4.2.4_81684/vboxhost/vboxdrv/r0drv/memobj-r0drv.c

comment:23 Changed 15 months ago by frank

Thanks. We now have confirmation that this is indeed a gcc bug, see the other comments in gcc bug  55940.

comment:24 Changed 15 months ago by frank

I have committed a workaround. Could you apply r44302 to your VirtualBox kernel modules in /usr/src/vboxhost-x.y.z/vboxdrv, recompile the kernel modules and verify if it works for you now?

comment:25 Changed 15 months ago by kimrhh

confirmed this works :) thanks a lot

comment:26 Changed 15 months ago by frank

  • Summary changed from failed to start kernel module vboxdrv (custom kernel) to failed to start kernel module vboxdrv (custom kernel) => Fixed in SVN

Thanks for the feedback. The workaround will be part of the next maintenance release.

comment:27 Changed 14 months ago by frank

  • Status changed from new to closed
  • Resolution set to fixed

The workaround is part of the 4.2.8 kernel module sources.

Note: See TracTickets for help on using tickets.

www.oracle.com
ContactPrivacy policyTerms of Use