Re: [Xen-devel] xl pci-attach silently fails the first time
On Wed, 2014-12-03 at 21:31 -0400, Konrad Rzeszutek Wilk wrote: so something is busted as you surmised I haven't been following closely, so this may be completely unrelated to the issue under discussion, but doesn't in-HVM-guest hotplug require a guest side driver to be loaded? I recall in the dim and distant past having to manually load some sort of foo_hp.ko. Ian. ___ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
Re: [Xen-devel] xl pci-attach silently fails the first time
On Dec 4, 2014 4:29 AM, Ian Campbell ian.campb...@citrix.com wrote: On Wed, 2014-12-03 at 21:31 -0400, Konrad Rzeszutek Wilk wrote: so something is busted as you surmised I haven't been following closely, so this may be completely unrelated to the issue under discussion, but doesn't in-HVM-guest hotplug require a guest side driver to be loaded? I recall in the dim and distant past having to manually load some sort of foo_hp.ko. The ACPI PCI hotplug machinery does all of that nicely. As mentioned - it works nicely - except that the BDF that is chosen is at an wrong slot. In the 'xend' days in recall this working nicely and sticking the PF at slot 5 and beyond. Though I have no clue how it dealt with multiple emulated NICs and such. Ian. ___ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
Re: [Xen-devel] xl pci-attach silently fails the first time
On Tue, Dec 02, Konrad Rzeszutek Wilk wrote: On Tue, Dec 02, 2014 at 04:46:52PM +0100, Olaf Hering wrote: On Mon, Dec 01, Konrad Rzeszutek Wilk wrote: That is odd - I see any device 'hot-plugged' being added at 00:05 and further. Does this by any chance depend on the guest?! I mean, how is the guest I doubt it. Something is different here, likely just the non-pvops guest kernel. notified that a PCI device is gone (by unplug)? Maybe the pvops case just happens to work because the unplug happens early, perhaps before PCI discovery?! ACPI hotplug. And it does work after PCI discovery. In a pvops kernel, is the emulated but unplugged PCI hardware still listed with lspci? Olaf ___ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
Re: [Xen-devel] xl pci-attach silently fails the first time
On Wed, Dec 03, Olaf Hering wrote: On Tue, Dec 02, Konrad Rzeszutek Wilk wrote: On Tue, Dec 02, 2014 at 04:46:52PM +0100, Olaf Hering wrote: On Mon, Dec 01, Konrad Rzeszutek Wilk wrote: ACPI hotplug. And it does work after PCI discovery. In a pvops kernel, is the emulated but unplugged PCI hardware still listed with lspci? It is not. So thats why it happens to work. So how would I trigger an ACPI hotplug event within qemus unplug code? Olaf ___ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
Re: [Xen-devel] xl pci-attach silently fails the first time
On Mon, Dec 01, Sander Eikelenboom wrote: Monday, December 1, 2014, 3:34:09 PM, you wrote: actually be a workaround for the double pci-attach bug. Don't know about that bug. You just replied to it. ;-) Olaf ___ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
Re: [Xen-devel] xl pci-attach silently fails the first time
On Mon, Dec 01, Konrad Rzeszutek Wilk wrote: That is odd - I see any device 'hot-plugged' being added at 00:05 and further. Does this by any chance depend on the guest?! I mean, how is the guest notified that a PCI device is gone (by unplug)? Maybe the pvops case just happens to work because the unplug happens early, perhaps before PCI discovery?! Olaf ___ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
Re: [Xen-devel] xl pci-attach silently fails the first time
On Tue, Dec 02, 2014 at 04:46:52PM +0100, Olaf Hering wrote: On Mon, Dec 01, Konrad Rzeszutek Wilk wrote: That is odd - I see any device 'hot-plugged' being added at 00:05 and further. Does this by any chance depend on the guest?! I mean, how is the guest I doubt it. notified that a PCI device is gone (by unplug)? Maybe the pvops case just happens to work because the unplug happens early, perhaps before PCI discovery?! ACPI hotplug. And it does work after PCI discovery. Olaf ___ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
[Xen-devel] xl pci-attach silently fails the first time
I tried to attach a PCI device (IGB Virtual Function) to a HVM guest. To actually get it assigned its required to run pci-attach/pci-detach/pci-attach because it does not show up right away. Did anyone noticed this bug already, is there a fix? There is no error reported in dom0 dmesg, xl dmesg or qemu logfiles. My domU.cfg looks like this: name=domU memory=512 builder=hvm vif=['',] disk=[ 'file:/SLE-12-Server-DVD-x86_64-GM-DVD1.iso,hda:cdrom' ] serial=pty # xl pci-assignable-add 01:10.0 # xl pci-assignable-list :01:10.0 # xl create -f domU.cfg # xl console domU ## lspci gives just emulated PCI devices ## detach from domU console # xl pci-attach domU :01:10.0 # xl pci-list domU Vdev Device 04.0 :01:10.0 # xl console domU ## lspci gives just emulated PCI devices ## detach from domU console # xl pci-detach domU :01:10.0 # xl pci-attach domU :01:10.0 # xl pci-list domU Vdev Device 04.0 :01:10.0 # xl console domU ## lspci shows now also the assigned host device Olaf ___ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
Re: [Xen-devel] xl pci-attach silently fails the first time
On Mon, Dec 01, Olaf Hering wrote: # xl pci-assignable-add 01:10.0 # xl pci-assignable-list :01:10.0 # xl create -f domU.cfg # xl console domU ## lspci gives just emulated PCI devices ttyS0:Rescue:~ # lspci 00:00.0 Host bridge: Intel Corporation 440FX - 82441FX PMC [Natoma] (rev 02) 00:01.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II] 00:01.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton II] 00:01.3 Bridge: Intel Corporation 82371AB/EB/MB PIIX4 ACPI (rev 03) 00:02.0 Unassigned class [ff80]: XenSource, Inc. Xen Platform Device (rev 01) 00:03.0 VGA compatible controller: Cirrus Logic GD 5446 00:04.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL-8100/8101L/8139 PCI Fast Ethernet Adapter (rev ff) ## detach from domU console # xl pci-attach domU :01:10.0 # xl pci-list domU Vdev Device 04.0 :01:10.0 # xl console domU ## lspci gives just emulated PCI devices ttyS0:Rescue:~ # lspci 00:00.0 Host bridge: Intel Corporation 440FX - 82441FX PMC [Natoma] (rev 02) 00:01.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II] 00:01.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton II] 00:01.3 Bridge: Intel Corporation 82371AB/EB/MB PIIX4 ACPI (rev 03) 00:02.0 Unassigned class [ff80]: XenSource, Inc. Xen Platform Device (rev 01) 00:03.0 VGA compatible controller: Cirrus Logic GD 5446 00:04.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL-8100/8101L/8139 PCI Fast Ethernet Adapter (rev 01) ## detach from domU console # xl pci-detach domU :01:10.0 Now lspci shows that the emulated network card is gone. ttyS0:Rescue:~ # lspci 00:00.0 Host bridge: Intel Corporation 440FX - 82441FX PMC [Natoma] (rev 02) 00:01.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II] 00:01.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton II] 00:01.3 Bridge: Intel Corporation 82371AB/EB/MB PIIX4 ACPI (rev 03) 00:02.0 Unassigned class [ff80]: XenSource, Inc. Xen Platform Device (rev 01) 00:03.0 VGA compatible controller: Cirrus Logic GD 5446 # xl pci-attach domU :01:10.0 # xl pci-list domU Vdev Device 04.0 :01:10.0 # xl console domU ## lspci shows now also the assigned host device So the actual bug is that the very first time after pci-attach the guests 00:04.0 PCI device is (most likely) replaced with the host PCI device. Just the guest does not notice that 00:04.0 was actually already gone after unplug. I wonder why the unplug code in xen_platform.c does just a qdev_free() instead of a real pci-detach kind of thing. I'm sure this could be done at least for emulated network cards. Olaf ___ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
Re: [Xen-devel] xl pci-attach silently fails the first time
On Mon, Dec 01, Ian Campbell wrote: On Mon, 2014-12-01 at 15:34 +0100, Olaf Hering wrote: On Mon, Dec 01, Sander Eikelenboom wrote: Hmm the wiki also still mentions it: http://wiki.xen.org/wiki/Xen_PCI_Passthrough It was the ability with xend + qemu-trad to be able to specify the slot to use in the guest for the pci device. See docs/misc/vtd.txt .. that seems it has never been updated :-) (together with passing through a multifunction devices as multifunction inside the guest this hasn't got implemented in neither libxl and qemu-xen.) So this looks like we a lost feature in libxl. See http://bugs.xenproject.org/xen/bug/22 (I think that's what that is) Yes, the virtual function part is covered by #22. Thanks. Olaf ___ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel