xend vs xl with pci=['<bdf'] wherein the '<bdf>' are not owned by pciback or pcistub will still launch.

xend vs xl with pci=['<bdf'] wherein the '<bdf>' are not owned by pciback or pcistub will still launch.

[Konrad Rzeszutek Wilk]

In Xend, if I had a pci entry in the guest config and the 
PCI device was not assigned to xen-pciback or pci-stub it
would refuse to launch the guest.

Not so with 'xl'. It will complain but still launch:

-bash-4.1# cd drivers/pciback/
-bash-4.1# ls
0000:01:00.0  0000:03:08.1  0000:03:0a.0  0000:03:0b.1       irq_handlers  new_slot    remove_id    uevent
0000:01:00.1  0000:03:09.0  0000:03:0a.1  bind               module        permissive  remove_slot  unbind
0000:03:08.0  0000:03:09.1  0000:03:0b.0  irq_handler_state  new_id        quirks      slots
-bash-4.1# echo "0000:03:0b.0" > unbind
-bash-4.1# echo "0000:03:0b.1" > unbind
-bash-4.1# xl create /mnt/lab/security/security.cfg  
Parsing config from /mnt/lab/security/security.cfg
libxl: error: libxl_pci.c:1055:libxl__device_pci_add: PCI device 0:3:b.0 is not assignable
libxl: error: libxl_pci.c:1055:libxl__device_pci_add: PCI device 0:3:b.1 is not assignable
-bash-4.1# xl list
Name                                        ID   Mem VCPUs      State   Time(s)
Domain-0                                     0  2047     4     r-----      14.7
security                                     1  1023     1     -b----       8.0
-bash-4.1# 
-bash-4.1# cat /mnt/lab/security/security.cfg |grep -v \#
device_model_version="qemu-xen-traditional"
builder="hvm"
memory = 1024
name = "security"
vcpus=1
vif = [ 'mac=00:0F:4B:00:00:84,bridge=switch' ]
disk = [ 'phy:/dev/sda,xvda,w' ]
pci= ['0000:03:08.0', '000:03:08.1', '0000:03:09.0', '0000:03:09.1', '0000:03:0a.0', '0000:03:0a.1', '0000:03:0b.0', '0000:03:0b.1']
vnc=1
vnclisten='0.0.0.0'
vncunused=1
serial="pty"


And naturally when shutting/destroying the guest it will say:
-bash-4.1# xl destroy 1
libxl: error: libxl_pci.c:1265:do_pci_remove: xc_deassign_device failed: No such device
libxl: error: libxl_pci.c:1265:do_pci_remove: xc_deassign_device failed: No such device

(XEN) [2014-01-06 17:54:39] deassign 0000:03:0b.0 from dom1 failed (-19)
(XEN) [2014-01-06 17:54:39] deassign 0000:03:0b.1 from dom1 failed (-19)

because it tries to de-allocate them even though they were not
part of the guest.

Re: xend vs xl with pci=['<bdf'] wherein the '<bdf>' are not owned by pciback or pcistub will still launch.

[Ian Campbell]

create ^
thanks

On Mon, 2014-01-06 at 12:57 -0500, Konrad Rzeszutek Wilk wrote:
> In Xend, if I had a pci entry in the guest config and the 
> PCI device was not assigned to xen-pciback or pci-stub it
> would refuse to launch the guest.
> 
> Not so with 'xl'. It will complain but still launch:

It looks like domcreate_attach_pci() is ignoring the result of
libxl__device_pci_add(). It appears to have always done so.

I suppose there is an argument that there are usecases where starting
the domain at all even without the full set of devices is better than
not starting it at all, but I think I agree that the default should be
to fail if some devices are not available.

Is this a blocker for you for 4.4 or can it wait for 4.5?

> 
> -bash-4.1# cd drivers/pciback/
> -bash-4.1# ls
> 0000:01:00.0  0000:03:08.1  0000:03:0a.0  0000:03:0b.1       irq_handlers  new_slot    remove_id    uevent
> 0000:01:00.1  0000:03:09.0  0000:03:0a.1  bind               module        permissive  remove_slot  unbind
> 0000:03:08.0  0000:03:09.1  0000:03:0b.0  irq_handler_state  new_id        quirks      slots
> -bash-4.1# echo "0000:03:0b.0" > unbind
> -bash-4.1# echo "0000:03:0b.1" > unbind
> -bash-4.1# xl create /mnt/lab/security/security.cfg  
> Parsing config from /mnt/lab/security/security.cfg
> libxl: error: libxl_pci.c:1055:libxl__device_pci_add: PCI device 0:3:b.0 is not assignable
> libxl: error: libxl_pci.c:1055:libxl__device_pci_add: PCI device 0:3:b.1 is not assignable
> -bash-4.1# xl list
> Name                                        ID   Mem VCPUs      State   Time(s)
> Domain-0                                     0  2047     4     r-----      14.7
> security                                     1  1023     1     -b----       8.0
> -bash-4.1# 
> -bash-4.1# cat /mnt/lab/security/security.cfg |grep -v \#
> device_model_version="qemu-xen-traditional"
> builder="hvm"
> memory = 1024
> name = "security"
> vcpus=1
> vif = [ 'mac=00:0F:4B:00:00:84,bridge=switch' ]
> disk = [ 'phy:/dev/sda,xvda,w' ]
> pci= ['0000:03:08.0', '000:03:08.1', '0000:03:09.0', '0000:03:09.1', '0000:03:0a.0', '0000:03:0a.1', '0000:03:0b.0', '0000:03:0b.1']
> vnc=1
> vnclisten='0.0.0.0'
> vncunused=1
> serial="pty"
> 
> 
> And naturally when shutting/destroying the guest it will say:
> -bash-4.1# xl destroy 1
> libxl: error: libxl_pci.c:1265:do_pci_remove: xc_deassign_device failed: No such device
> libxl: error: libxl_pci.c:1265:do_pci_remove: xc_deassign_device failed: No such device
> 
> (XEN) [2014-01-06 17:54:39] deassign 0000:03:0b.0 from dom1 failed (-19)
> (XEN) [2014-01-06 17:54:39] deassign 0000:03:0b.1 from dom1 failed (-19)
> 
> because it tries to de-allocate them even though they were not
> part of the guest.
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel

Processed: Re: xend vs xl with pci=['<bdf'] wherein the '<bdf>' are not owned by pciback or pcistub will still launch.

[xen]

Processing commands for xen@bugs.xenproject.org:

> create ^
Created new bug #27 rooted at `<20140106175713.GB28636@phenom.dumpdata.com>'
Title: `Re: [Xen-devel] xend vs xl with pci=['<bdf'] wherein the '<bdf>' are not owned by pciback or pcistub will still launch.'
> thanks
Finished processing.

Modified/created Bugs:
 - 27: http://bugs.xenproject.org/xen/bug/27 (new)

---
Xen Hypervisor Bug Tracker
See http://wiki.xen.org/wiki/Reporting_Bugs_against_Xen for information on reporting bugs
Contact xen-bugs-owner@bugs.xenproject.org with any infrastructure issues

Re: xend vs xl with pci=['<bdf'] wherein the '<bdf>' are not owned by pciback or pcistub will still launch.

[Konrad Rzeszutek Wilk]

On Tue, Jan 07, 2014 at 11:22:58AM +0000, Ian Campbell wrote:
> create ^
> thanks
> 
> On Mon, 2014-01-06 at 12:57 -0500, Konrad Rzeszutek Wilk wrote:
> > In Xend, if I had a pci entry in the guest config and the 
> > PCI device was not assigned to xen-pciback or pci-stub it
> > would refuse to launch the guest.
> > 
> > Not so with 'xl'. It will complain but still launch:
> 
> It looks like domcreate_attach_pci() is ignoring the result of
> libxl__device_pci_add(). It appears to have always done so.
> 
> I suppose there is an argument that there are usecases where starting
> the domain at all even without the full set of devices is better than
> not starting it at all, but I think I agree that the default should be
> to fail if some devices are not available.

<nods> The guest wasn't too happy about some of them missing :-)

> 
> Is this a blocker for you for 4.4 or can it wait for 4.5?

Not a blocker. It can wait, just want to make sure we don't forget.