linux-kernel - Re: 3.19-rc4: Xen pci-passthrough regression, bisected to commit cffe0a2b5a34c95a4dadc9ec7132690a5b0f6687 "x86, irq: Keep balance of IOAPIC pin reference count"

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Wed, 14 Jan 2015 17:17:36 +0100
From:	Sander Eikelenboom <linux@...elenboom.it>
To:	David Vrabel <david.vrabel@...rix.com>
CC:	Jiang Liu <jiang.liu@...ux.intel.com>,
	Konrad Rzeszutek Wilk <konrad.wilk@...cle.com>,
	<xen-devel@...ts.xenproject.org>, <linux-kernel@...r.kernel.org>
Subject: Re: 3.19-rc4: Xen pci-passthrough regression, bisected to commit cffe0a2b5a34c95a4dadc9ec7132690a5b0f6687 "x86, irq: Keep balance of IOAPIC pin reference count"


Wednesday, January 14, 2015, 3:58:33 PM, you wrote:

> On 14/01/15 14:15, Sander Eikelenboom wrote:
>> Hi Gerry / David / Konrad,
>> 
>> Some more testing uncovered another issue under Xen, this time with PCI-passthrough.

> What device?  In particular what interrupts is it using?

Hi David,

Here is a more complete set of debug logs, for both with and without the revert.
- dmesg
- xl-dmesg with output of debug keys 'i, M, z'
- lspci part of the two devices from the guest
- /proc/interrupts

The wifi NIC (dom0: 02:00.0 guest: 00:05.0) uses legacy interrupts and gives troubles:
It's using:
        Interrupt: pin A routed to IRQ 36
        Capabilities: [50] MSI: Enable- Count=1/1 Maskable- 64bit-
        36:      14413  xen-pirq-ioapic-level  ath9k

The other NIC (dom0: 00:19.0 guest: 00:06.0) uses MSI interrupts and that works fine:
        Interrupt: pin A routed to IRQ 57
        Capabilities: [d0] MSI: Enable+ Count=1/1 Maskable- 64bit+
        57:        182  xen-pirq-msi       eth0
--
Sander  

>> I have bisected it to the following commit: 
>> cffe0a2b5a34c95a4dadc9ec7132690a5b0f6687 "x86, irq: Keep balance of IOAPIC pin reference count"
>> 
>> It causes these symptoms:
>> 
>> - On Intel
>>   - Running on Xen with pci devices seized on host boot with xen-pciback.hide= parameter
>>   - Running a HVM guest with PCI passthrough of two devices (NIC + wireless NIC)
>>   - While the driver loads fine, the device isn't working properly, looking in /proc/interrupts in the guest
>>     shows that it doesn't receive any interrupts.
>>   - Reverting this particular commit (in the dom0 kernel only) makes the device receive interrupts and work properly again.
>> 
>> - On AMD (more subtle symptom) 
>>   - Running on Xen with pci devices seized on host boot with xen-pciback.hide= parameter
>>   - Running a HVM guest with PCI passthrough of one devices (videograbber)
>>   - While the driver loads fine and the device looks like it's working, the videostream isn't stable and it skips or repeats frames.
>>   - Reverting this particular commit (in the dom0 kernel only) makes the device work properly again with a stable videostream.
>> 
>> --
>> Sander
>> 


View attachment "dmesg.txt" of type "text/plain" (54485 bytes)

View attachment "dmesg-revert.txt" of type "text/plain" (54459 bytes)

View attachment "lspci-guest.txt" of type "text/plain" (3357 bytes)

View attachment "lspci-guest-revert.txt" of type "text/plain" (3357 bytes)

View attachment "proc-interrupts.txt" of type "text/plain" (1601 bytes)

View attachment "proc-interrupts-revert.txt" of type "text/plain" (1601 bytes)

View attachment "xl-dmesg.txt" of type "text/plain" (45156 bytes)

View attachment "xl-dmesg-revert.txt" of type "text/plain" (45195 bytes)