Re: [Xen-users] PCI Passthrough without VT-d

["Sergio Charpinel Jr." <sergiocharpinel@xxxxxxxxx>]

Hi,

I created a PV guest, and did a passthrough on my PCI USB Controller, to use the webcam.

But, I'm receiving a kernel panic message on my domU afte boot.

I'm using CentOS 5.4 and Xen 3.4.2

Password: Fatal DMA error! Please use 'swiotlb=force'

----------- [cut here ] --------- [please bite here ] ---------

Kernel BUG at arch/x86_64/kernel/../../i386/kernel/pci-dma-xen.c:394

invalid opcode: 0000 [1] SMP 

last sysfs file: /devices/xen/pci-0/pci0000:00/0000:00:00.3/usb1/1-1/1-1:1.0/usbdev1.2_ep82/dev

CPU 0 

Modules linked in: autofs4 hidp rfcomm l2cap bluetooth lockd sunrpc ip_conntrack_netbios_ns ipt_MASQUERADE xt_mark iptable_nat ip_nat xt_MARK iptable_mangle ipt_REJECT xt_state ip_conntrack nfnetlink iptable_filter ip_tables ip6t_REJECT xt_tcpudp ip6table_filter ip6_tables x_tables ipv6 xfrm_nalgo crypto_api dm_mirror dm_multipath scsi_dh scsi_mod parport_pc lp parport stv680 compat_ioctl32 videodev v4l1_compat pcspkr v4l2_common xennet dm_raid45 dm_message dm_region_hash dm_log dm_mod dm_mem_cache xenblk ext3 jbd uhci_hcd ohci_hcd ehci_hcd

Pid: 1752, comm: zmc Not tainted 2.6.18-164.11.1.el5xen #1

RIP: e030:[<ffffffff8027217b>]  [<ffffffff8027217b>] dma_map_single+0x12d/0x17d

RSP: e02b:ffff88000cfb1c78  EFLAGS: 00010282

RAX: 000000000000002f RBX: ffff88000e860000 RCX: ffffffff804f3ba8

RDX: 0000000000000000 RSI: 0000000000000001 RDI: 0000000000000000

RBP: 0000000035032000 R08: ffffffff804f3ba8 R09: 0000000000000000

R10: 000000000000002d R11: ffff88001f4230c0 R12: 0000000000019610

R13: ffff88001fdef870 R14: ffff88001fb5f980 R15: ffff880010296ad4

FS:  00002b8745423c10(0000) GS:ffffffff805ca000(0000) knlGS:0000000000000000

CS:  e033 DS: 0000 ES: 0000

Process zmc (pid: 1752, threadinfo ffff88000cfb0000, task ffff88000f4530c0)

Stack:  ffff88000f4530c0  ffff880010296ac0  00000000ffffffff  ffff880010296ac0 

 ffff88001ff7d400  ffffffff803e4200  0000000000000000  000000d0802572b0 

 fffffffffff7ffff  ffffffff802c2e18 

Call Trace:

 [<ffffffff803e4200>] hcd_submit_urb+0x693/0x748

 [<ffffffff802c2e18>] mod_zone_page_state+0x27/0x3d

 [<ffffffff8020b242>] kmem_cache_alloc+0x62/0x6d

 [<ffffffff8025e6e3>] cache_alloc_refill+0x13c/0x4e5

 [<ffffffff802cfc34>] __kmalloc+0x8f/0x9f

 [<ffffffff881377e0>] :stv680:stv680_start_stream+0x18b/0x1c5

 [<ffffffff881397a0>] :stv680:stv680_do_ioctl+0x4e1/0x57b

 [<ffffffff8811ddda>] :videodev:video_usercopy+0x1a1/0x265

 [<ffffffff881392bf>] :stv680:stv680_do_ioctl+0x0/0x57b

 [<ffffffff8031a360>] inode_has_perm+0x56/0x63

 [<ffffffff80222789>] __up_read+0x19/0x7f

 [<ffffffff802676cd>] do_page_fault+0xfae/0x12e0

 [<ffffffff8026ef47>] monotonic_clock+0x35/0x7b

 [<ffffffff80243f5c>] do_ioctl+0x55/0x6b

 [<ffffffff802316b5>] vfs_ioctl+0x457/0x4b9

 [<ffffffff8024e68e>] sys_ioctl+0x59/0x78

 [<ffffffff802602f9>] tracesys+0xab/0xb6

Code: 0f 0b 68 3a 2b 49 80 c2 8a 01 4d 85 ed 74 11 49 8b 85 18 02 

RIP  [<ffffffff8027217b>] dma_map_single+0x12d/0x17d

 RSP <ffff88000cfb1c78>

 <0>Kernel panic - not syncing: Fatal exception

2010/3/2 Jan Češčut <Jan.Cescut@xxxxxx>

Thanks for thorough explanation.

Have a nice day,
Jan

-----Original Message-----
From: Pasi Kärkkäinen [mailto:pasik@xxxxxx]
Sent: 27. februar 2010 14:04
To: Jan Češčut
Cc: xen-users@xxxxxxxxxxxxxxxxxxx
Subject: Re: [Xen-users] PCI Passthrough without VT-d

On Fri, Feb 26, 2010 at 11:29:22PM +0100, Jan ?eš?ut wrote:
>    As I read XEN supports assigning a pci device to an unprivileged domain
>    without hardware supporting it. Has anyone already tried it? Are there any
>    security risks? If I understand correctly how PCI passthrough works the
>    performance should be the same as using the pci device in native mode. Is
>    it so? I have a PCI video card which would like to use inside a VM running
>    Windows XP.
>

Xen supports PCI passthrough to _PV_ (paravirtual) guests without VT-d,
and has actually supported that for years. There are some potential security
risks in this, since the PV guest gets full DMA control of the PCI device
and could use it for malicious purposes.

Xen PCI passthrough to HVM guests (=Windows) requires VT-d hardware support.

Also, PCI passthrough of a VGA/video card is not as simple as PCI passthrough
of other cards (nic, disk controller, usb controller).

VGA has lots of legacy stuff related to it, some memory ranges, IO ports, VGA BIOS,
etc that have to be 'passed through' aswell, and emulated.

Xen 4.0.0 will have PCI passthrough support of primary VGA adapters, but it requires
VT-d support as stated already earlier.

-- Pasi

ps. There is actually a hack/patch available that allows PCI passthrough to HVM guest
without VT-d, but that only works for the _first_ started HVM guest, and it's experimental
and not supported in any way. iirc the patch is available in xen-devel archives.

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users

--
Sergio Roberto Charpinel Jr.

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users