[vbox-dev] Chipset support for Message Signaled Interrupts (MSI)

Thu Mar 31 03:31:26 GMT 2016

Thanks for your reply.

As I mentioned, I already know that this device does not support MSI.  That is why I need to use legacy IRQ for it.  There are no extended config capabilities at all in this device.  I thought you meant the driver may be expecting MSI and therefore does not support IRQ, and that is what I don’t know how to investigate.

As a separate issue, I also have a second device which does require MSI support, and does have a proper MSI struct in the extended PCI Config.  That device is driving me towards using ICH9 instead of PIIX3.  I have moved that discussion to a different email thread, which is currently pending moderator approval due to large attachment.

But I need to get both devices working in the same VM.  So if the second one requires that I use ICH9, then I have to figure out why the first one cannot get legacy IRQ through.  I will start drilling down from PDMDevHlpPCISetIrq()…

//  RicV

From: Ananth Pallapothu [mailto:apallapothu at gmail.com]
Sent: Wednesday, March 30, 2016 19:57
To: Vilbig, Ric
Cc: vbox-dev at virtualbox.org
Subject: Re: [vbox-dev] Chipset support for Message Signaled Interrupts (MSI)

You could check MSI enablement by running down config space of the device. Check the capabilities pointer and track down MSI capability.

I am assuming you have added MSI registration code in your device model matching the end point device.
Past that, you simply need to look out for any upstream memory writes @ address 0xFEEXXXXX, upon receiving, simply invoke PDMDevHlpPCISetIrq(). As long as the model is MSI enabled code will pick appropriate MSI related functions.

I am not sure I can comment more on Legacy without actually knowing, debugging through the code.

On Wed, Mar 30, 2016 at 6:33 PM, Vilbig, Ric <ric_vilbig at mentor.com<mailto:ric_vilbig at mentor.com>> wrote:
Hi,

I am not sure if the drivers are MSI enabled, or how to check that — I just installed the guest from standard iso distribution.  But I should think when the driver is bound to a new device, it would check the capabilities of that device.  This particular device (USB mass storage controller) does not support MSI.

Yes, I am calling PDMDevHlpPCISetIrq() to signal the interrupt.  This same virtual device model works well in a VM that uses the PIIX3 chipset, but I need to move to the ICH9 chipset to support a different device (PCIe switch) which does require MSI support.  That is actually my goal right now, I just thought moving the older device that already works with PIIX3 over to ICH9 would be a good first step in that transition.  In the end I need both of these devices working in the same VM.

//  RicV

From: Ananth Pallapothu [mailto:apallapothu at gmail.com<mailto:apallapothu at gmail.com>]
Sent: Saturday, March 26, 2016 11:00
To: Vilbig, Ric
Cc: vbox-dev at virtualbox.org<mailto:vbox-dev at virtualbox.org>

Subject: Re: [vbox-dev] Chipset support for Message Signaled Interrupts (MSI)

Are you sure your drivers aren't MSI enabled ? Not any known issues with Legacy on ICH9 or I should rather say I didn't have any issue using Legacy interrupts.

I am assuming you invoked PDM*irq function with your device appropriately (upon receiving the interrupt) ?

On Thu, Mar 24, 2016 at 7:11 PM, Vilbig, Ric <ric_vilbig at mentor.com<mailto:ric_vilbig at mentor.com>> wrote:
Now I have the opposite question.  Are IRQ type interrupts (as in PCI Config 0x3d IntPin) expected to work with ICH9?

I have a virtual PCI device which works with PIIX3 quite nicely using IRQ4 — a USB3 mass storage device.  But when plugged into the ICH9 VM the device is not detected within the guest, using IRQ1/2/3/4.

The guest must know the device is present, because it does lots of MMIO Rd/Wr to it, and then the device does some DMA, and raises an interrupt.  Everything looks quite similar between the ICH9 and PIIX3 case, except the former does nothing more after raising the interrupt, whereas the latter continues working with lots more DMA, interrupts, and a usable mass storage device in the guest.

From what I have found so far, the interrupt is passed down into PDM, but does not actually do anything, and dmesg in the guest says "can't derive routing for PCI INT n".

Thanks,

//  RicV

From: Ananth Pallapothu [mailto:apallapothu at gmail.com<mailto:apallapothu at gmail.com>]
Sent: Monday, March 21, 2016 15:55
To: Vilbig, Ric
Cc: VirtualBox developer's list
Subject: Re: [vbox-dev] Chipset support for Message Signaled Interrupts (MSI)

MSI's came into existence after PIIX3.

Yes, if you need MSI you need to use ICH9.
PIIX3 doesn't support MSI
On Mar 21, 2016 6:43 PM, "Vilbig, Ric" <ric_vilbig at mentor.com<mailto:ric_vilbig at mentor.com>> wrote:
Hi Experts,

I posted the following on the VIrtualBox forum, but Perry recommended I contact you by the email list.
https://forums.virtualbox.org/viewtopic.php?f=10&t=76867

________________________________
I am working on a custom PCI device model, plugged into VirtualBox via PDM. It's mostly working, lspci for example shows it's present. But I am having a problem with Message Signaled Interrupts (MSI).

The following quote from Ch.3 of the VirtualBox User Manual implies that the ICP9 chipset must be used for MSI support, and I have seen supporting evidence in web searches and VBox source code (OSE 5.0.16 tarball).

“...VirtualBox 4.0 introduced an emulation of the more modern ICH9 chipset, which supports PCI express,
three PCI buses, PCI-to-PCI bridges and Message Signaled Interrupts (MSI)...”

Can someone confirm or refute with certainty that the PIIX3 chipset does not support MSI and ICH9 is required for device models using MSI?

How worried should I be about this excerpt from the same paragraph? A quick test of ICH9 VM did not go well. I've actually tried with Fedora 23, Fedora 18 and Ubuntu 14 guests (all 64-bit) without success, but not yet dug into it because this quote worries me.

“Note that the ICH9 support is experimental and not recommended for guest operating systems which do not require it.”
________________________________

Thanks in advance for your help.

_____________________________________________

Ric Vilbig
Mentor Graphics, Emulation Division
46871 Bayside Parkway, Fremont CA, 94538
Phone:  510-354-7360<tel:510-354-7360>
Mobile: 408-529-2365<tel:408-529-2365>
email:  ric_vilbig at mentor.com<mailto:ric_vilbig at mentor.com>

_______________________________________________
vbox-dev mailing list
vbox-dev at virtualbox.org<mailto:vbox-dev at virtualbox.org>
https://www.virtualbox.org/mailman/listinfo/vbox-dev

_______________________________________________
vbox-dev mailing list
vbox-dev at virtualbox.org<mailto:vbox-dev at virtualbox.org>
https://www.virtualbox.org/mailman/listinfo/vbox-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.virtualbox.org/pipermail/vbox-dev/attachments/20160331/8e39e3a5/attachment.html>