Re: Some MSI related bugs when trying to use VT-d. Help identify software vs hardware problem?

[Andrew Bobulsky <rulerof@gmail.com>]

[-- Attachment #1.1: Type: text/plain, Size: 13415 bytes --]

Hello Konrad!

On May 22, 2013, at 1:21 PM, Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
wrote:

On Fri, May 10, 2013 at 11:20:28AM -0400, Andrew Bobulsky wrote:

Hello Konrad,


Thanks for getting back to me!


Sure, sorry for the long response, vacation and other things have delayed
my response time.


In a sheer twist of irony, I'm now the one who is on vacation :D

I'll be returning home on the 28th, and will be able to answer with better
detail at that point, but I can give you a few answers in the meantime.
 Here goes!




On May 10, 2013, at 9:36 AM, Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>

wrote:


On Wed, May 08, 2013 at 09:45:14AM +0100, George Dunlap wrote:


On Tue, May 7, 2013 at 11:38 AM, Andrew Bobulsky <rulerof@gmail.com> wrote:


Hello List!



I'm having another [rather fruitless] go at trying to get PCIe passthrough


to work on my Radeon 6990 card(s).  I have an i7 920 chip in a Gigabyte


GA-EX58-EXTREME board that I've flashed a modded BIOS into to add VT-d


support... I found the BIOS image on a BIOS modding forum maybe a year or


two ago.




I stuck an extra Highpoint RocketU 1144A USB 3 card into the board, because


I know it works *very well* with IOMMU and its architecture is really


convenient... each port on the back is essentially its own PCIe device[1].


I was able to "xl pci-assignable-add" the usb controllers, and attach and


detach them at will to a Server 2012 DomU.  The dmesg output from that event


looked like this:



[49857.921550] xhci_hcd 0000:06:00.0: remove, state 4


[49857.921555] usb usb15: USB disconnect, device number 1


[49857.921600] xHCI xhci_drop_endpoint called for root hub


[49857.921602] xHCI xhci_check_bandwidth called for root hub


[49857.921681] xhci_hcd 0000:06:00.0: USB bus 15 deregistered


[49857.921686] xhci_hcd 0000:06:00.0: remove, state 1


[49857.921689] usb usb13: USB disconnect, device number 1


[49857.921691] usb 13-1: USB disconnect, device number 4


[49857.953278] xHCI xhci_drop_endpoint called for root hub


[49857.953280] xHCI xhci_check_bandwidth called for root hub


[49857.963158] xhci_hcd 0000:06:00.0: USB bus 13 deregistered


[49857.963455] pciback 0000:06:00.0: seizing device


[49857.963500] xen: registering gsi 17 triggering 0 polarity 1


[49857.963503] Already setup the GSI :17


[49857.963514] pciback 0000:06:00.0: MSI-X preparation failed (-38)




Nonetheless, it works quite well!  I played audio through a USB headset from


the DomU to confirm it as well.





However, when I try to "xl pci-assignable-add" one of my VGA controllers


from the Radeon, the action completes, and "xl pci-assignable-list" shows


the device as available, but "xl pci-attach" never completes, and attemping


to "xl pci-assignable-add" the HDMI audo device never returns to the CLI


either.



OK, and is 0e:00.0 your VGA controller?


How do you assign the VGA controller? Do you do:


echo "0000:0e.00.0" > /sys/../radeon/unbind

echo "0000:0e.00.0" > /sys/../pciback/new_slot

echo "0000:0e.00.0" > /sys/../pciback/bind

?



I tried both the sysfs methods suggested by the wiki, as well as trying to

hide the device via grub. Pciback is compiled into the kernel (I can't say

I verified it, but it's definitely loaded and doesn't show up on lsmod :P).


I didn't check my dmesg output when using sysfs, as I just figured I was

doing something wrong as certain objects either didn't exist or wouldn't

accept input via the ">" operator.  I can't remember precisely, but I'd be

willing to go back and find out if you like.


I basically ended up doing "xl pci-assignable-add" to all of the functions

I needed.  In the case of the 6990, there are four, which for me are

0:d:0.0-1, and 0:e:0.0-1.  Two VGA controllers and their respective HDMI

audio devices.


There are two of these cards in the system, a total of 8 functions: four

VGA controllers and four HDMI audio devices.  The four listed here are on

my "unused" card.  The AMD CCC does detect all four devices, so I assume

that the driver is bound to all of them by the time I get around to running

my xl commands.



There is no visible output in dmesg from the attempt on the HDMI audio


device, but when I run it against the VGA controller, I get this:



[55817.715309] pciback 0000:0e:00.0: seizing device


[55817.737444] ------------[ cut here ]------------


[55817.737447] kernel BUG at drivers/pci/msi.c:346!


[55817.737449] invalid opcode: 0000 [#1] PREEMPT SMP


[55817.737451] Modules linked in: xt_physdev iptable_filter ip_tables


x_tables tun parport_pc ppdev lp parport bnep rfcomm bluetooth rfkill crc16


cpufreq_stats binfmt_misc fuse bridge stp llc ext2 loop snd_hda_codec_hdmi


snd_hda_codec_realtek joydev mperf coretemp crc32c_intel fglrx(PO)


snd_hda_intel snd_hda_codec snd_usb_audio microcode hid_generic mxm_wmi


snd_usbmidi_lib evdev psmouse snd_seq_midi snd_seq_midi_event i2c_i801


pcspkr tpm_tis snd_hwdep tpm snd_rawmidi serio_raw snd_pcm i2c_core tpm_bios


snd_seq snd_timer snd_seq_device lpc_ich mfd_core snd ehci_pci soundcore


snd_page_alloc wmi xhci_hcd button processor thermal_sys sg sr_mod cdrom


ext3 jbd mbcache dm_mirror dm_region_hash dm_log dm_mod sd_mod crc_t10dif


usb_storage usbhid hid ahci libahci uhci_hcd ehci_hcd usbcore usb_common


e1000e


[55817.737482] CPU 6


[55817.737484] Pid: 18055, comm: xl Tainted: P           O 3.8.11 #1


Gigabyte Technology Co., Ltd. EX58-EXTREME/EX58-EXTREME


[55817.737485] RIP: e030:[<ffffffff811eab09>]  [<ffffffff811eab09>]


free_msi_irqs+0x5d/0x11b


[55817.737490] RSP: e02b:ffff88034211dd08  EFLAGS: 00010282


[55817.737491] RAX: ffff880420bbd600 RBX: ffff88042096fa80 RCX:


0000000000000000


[55817.737492] RDX: 0000000000000000 RSI: 0000000000000091 RDI:


0000000000000011


[55817.737493] RBP: ffff880421f81000 R08: ffff88042096fa80 R09:


ffff88034211dce4


[55817.737494] R10: ffff88034211dd16 R11: 0000000000000000 R12:


ffff880421f81858


[55817.737495] R13: 0000000000000001 R14: 0000000000000000 R15:


0000000000000001


[55817.737498] FS:  00007fa609475740(0000) GS:ffff88043a2c0000(0000)


knlGS:0000000000000000


[55817.737499] CS:  e033 DS: 0000 ES: 0000 CR0: 000000008005003b


[55817.737500] CR2: ffffffffff600400 CR3: 000000033a673000 CR4:


0000000000002660


[55817.737501] DR0: 0000000000000000 DR1: 0000000000000000 DR2:


0000000000000000


[55817.737503] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7:


0000000000000400


[55817.737504] Process xl (pid: 18055, threadinfo ffff88034211c000, task


ffff8804217b5ca0)


[55817.737504] Stack:


[55817.737505]  00000000000000a2 ffff880421f81000 0000000000000000


ffff8803ca2ab6c0


[55817.737507]  ffff880421f81098 ffff880421f810f8 ffff8803ca2abc00


ffffffff811eb21d


[55817.737509]  ffff880421f81000 ffffffff81240822 ffff880421f81098


ffff880421f81000


[55817.737510] Call Trace:


[55817.737513]  [<ffffffff811eb21d>] ? pci_disable_msi+0x28/0x41


[55817.737516]  [<ffffffff81240822>] ? xen_pcibk_reset_device+0x3a/0xa0


[55817.737518]  [<ffffffff8123fd9f>] ? pcistub_init_device+0x167/0x19c


[55817.737521]  [<ffffffff81108661>] ? __kmalloc+0xd6/0xe2


[55817.737523]  [<ffffffff8123ff08>] ? pcistub_probe+0x134/0x1b8


[55817.737525]  [<ffffffff811df0f5>] ? local_pci_probe+0x37/0x5d


[55817.737527]  [<ffffffff811dff7c>] ? pci_device_probe+0xc2/0xe3


[55817.737529]  [<ffffffff81277d71>] ? driver_probe_device+0xa1/0x1ac


[55817.737532]  [<ffffffff81276e38>] ? driver_bind+0x7e/0xc7


[55817.737534]  [<ffffffff81165859>] ? sysfs_write_file+0xd3/0x10f


[55817.737537]  [<ffffffff8110fb7d>] ? vfs_write+0xa4/0xfe


[55817.737539]  [<ffffffff813c22e5>] ? _raw_spin_lock+0xe/0x2a


[55817.737541]  [<ffffffff8110fcc8>] ? sys_write+0x58/0x92


[55817.737543]  [<ffffffff813c7f29>] ? system_call_fastpath+0x16/0x1b


[55817.737544] Code: 8a 3b 45 31 f6 41 d0 ef 44 89 f9 45 89 ef 83 e1 07 41


d3 e7 eb 1c 8b 7b 0c 44 01 f7 e8 59 74 eb ff 48 83 b8 90 00 00 00 00 74 04


<0f> 0b eb fe 41 ff c6 45 39 fe 7c df 48 8b 5b >10 48 83 eb 10 48


[55817.737560] RIP  [<ffffffff811eab09>] free_msi_irqs+0x5d/0x11b


[55817.737562]  RSP <ffff88034211dd08>


[55817.737563] ---[ end trace e1c5a8a903358804 ]---




Any chance anyone could help me identify the source of this problem?  Can I


work around it with software, or do I need a different motherboard or video


card to make it work?



cc'ing Konrad and a couple of other people who might be able to take a


look at the BUG



That looks to be:


344 #ifdef CONFIG_GENERIC_HARDIRQS


345                 for (i = 0; i < nvec; i++)


346                         BUG_ON(irq_has_action(entry->irq + i));


347 #endif


I have to say I hadn't actually compiled the kernel with GENERIC_HARDIRQS

in a while.


I take it back, it looks as if my kernel has been compiling with that
option.


Looking at the code the issue seems that the MSI is enabled when

the PCI device was assigned to xen-pciback. That looks like a bug

in the radeon driver.


And based on our config (thanks!) you could also do this on your Linux

command line:


xen-pciback.hide=(0e:00.0)


If you do that and try to pass in the radeon device does it work?



Strangely, this was actually the second or third thing I tried... It's like

my system is ignoring it for some reason.  I hid *all* of the devices I

wanted to use via grub... But they weren't hidden.  I don't recall if I

tried it with my USB 3 controller, as *it* worked just fine when yanking it

from Dom0 via xl.


What do you mean by hidden? Are you thinking that by doing 'lspci' you
wouldn't
see them? They will be visible in your dom0. It is just that the driver
(radeon) won't bind to them. You will see something like this (look for
pciback):

01:00.0 Ethernet controller: Intel Corporation 82571EB Gigabit Ethernet
Controller (rev 06)
       Subsystem: Intel Corporation PRO/1000 PT Dual Port Server Adapter
       Flags: fast devsel, IRQ 16
       Memory at fe4a0000 (32-bit, non-prefetchable) [disabled] [size=128K]
       Memory at fe480000 (32-bit, non-prefetchable) [disabled] [size=128K]
       I/O ports at e020 [disabled] [size=32]
       Expansion ROM at fe460000 [disabled] [size=128K]
       Capabilities: [c8] Power Management version 2
       Capabilities: [d0] MSI: Enable- Count=1/1 Maskable- 64bit+
       Capabilities: [e0] Express Endpoint, MSI 00
       Capabilities: [100] Advanced Error Reporting
       Capabilities: [140] Device Serial Number 00-15-17-ff-ff-8f-18-a2
======> Kernel driver in use: pciback <====================
       Kernel modules: e1000e


When I say that they weren't hidden, while I didn't check with lspci -v,
running xl pci-assignable-list shows that nothing is available to pass to a
DomU.  I have to xl pci-assignable-add every device I want to use
regardless of whether or not it is hidden via my GRUB command line.
 Additionally, the Radeon driver was bound to the card I hid via grub, and
its HDMI audio device was also bound to snd-hda-intel (I think) in spite of
it also being in the grub command line.

xl itself mentions that it is seizing devices from others drivers in order
to bind them to pciback.



I ended up on kernel 3.8.11 because various other kernels I tried,

including 3.4.9 and something from the 3.7.x line both resulted in X

failing to start, but only when booting Xen.  After playing kernel shuffle


X failing to start?


Blinking (or solid, I cant recall precisely) underscore-type cursor in the
upper left corner of the screen, and it appears to be at a BIOS-like
text-mode resolution.

What card at that point are you using? Is the radeon
or another?  How does it fail?


I tried with the Radeon 6990, and also with a Radeon 5850.  I tried with
and without the Radeon driver, too, it didn't seem to make much difference
:(

If you boot with 'drm.debug=255 debug loglevel=8' on your
command line what does dmesg (or /var/log/messages) show? Can you attach
that
please.


I usually rebooted the system so my dmesg was out of date by the time i get
my hands on it.  I could try to SSH in, which should work as the system
responds to Ctrl-Alt-Del by rebooting after a few seconds, or grab a
historical log....  I'll report back next week :)


Perhaps you can also attach the lspci so I can get an idea of what
is what in your box.


The best I have access to is an lspci -vtQ that I ran a while back:
http://pastebin.com/raw.php?i=4dGmneYi



for a while, I've landed, late last night, on 3.4.44, and it actually

resolved my MSI problem!


Funny enough, every kernel I've tried has been built with the config I

posted too, 3.4.44 included!  If I was just "doing it wrong" with the

3.8.11 kernel, then the solution of "use 3.4.44 instead" is acceptable for

me, but if I've truly hit on some kind of kernel bug that bears resolving,

I still have the dpkg images that I built of all the kernels in question

and can collect more data as needed.  I'm rebooting a lot right now anyway.

;)


Still, I'm troubled by my inability to hide devices via grub.  I just

expected it to work and am flabbergasted that it won't...  I'll send a copy

of my grub config if you like, but I don't want to pollute the developers

list with an issue that's probably more appropriate for the users list. :)


cat /proc/cmdline and 'xl info' would help.


Thanks so much for your time and help, Konrad!  I really, really appreciate

it!


Sure thing.


Cheers,
Andrew

[-- Attachment #1.2: Type: text/html, Size: 33491 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel