Re: [vfio-users] intel_iommu=on and aacraid / Adaptec 3805
the command "dmesg | grep -i dma" Give no results. Is there anything else i should check Alex? If the solution to this is to submit a bug report, who should i send it too? I believe the AACRAID driver is built into the Linux kernel. On Tue, Jul 12, 2016 at 9:51 AM, Alex Williamsonwrote: > On Sun, 10 Jul 2016 13:17:41 -0500 > David wrote: > >> I have run into a strange problem, while setting up my Fedora 24 box for >> KVM, i noticed that my raid array stopped showing as an available drive. >> After a lot of troubleshooting and reinstalling linux to this PC 2 more >> times, i have narrowed it down to one setting in my GRUB config. >> intel_iommu=on >> >> Just taking that one setting out and rebuilding my grub2-efi.cfg will make >> the array readable again. When IOMMU is on, the system can see that there >> is a raid card and array, but it lists the partition table as unknown. It >> also will not successfully create a new GPT partition table on the array. >> With IOMMU off, it can read the partition table and partition fine, read >> and write data, and everything works fine. >> >> My raid is 4 1TB disks in Raid 10e, GPT partition formatted NTFS. >> >> $ lspci -v -s 03:0e.0 >> 03:0e.0 RAID bus controller: Adaptec AAC-RAID >> Subsystem: Adaptec 3805 >> Flags: bus master, stepping, 66MHz, medium devsel, latency 32, IRQ 57 >> Memory at fa60 (64-bit, non-prefetchable) [size=2M] >> Expansion ROM at fa80 [disabled] [size=256K] >> Capabilities: >> Kernel driver in use: aacraid >> Kernel modules: aacraid > > > No DMAR faults in dmesg? There's a fair chance this is one of those > devices that does DMA with the wrong requester ID and therefore > enabling the IOMMU prevents it from working. We have a mechanism in > the kernel to handle such DMA aliasing if you can verify the issue. > Thanks, > > Alex -- David david...@gmail.com ___ vfio-users mailing list vfio-users@redhat.com https://www.redhat.com/mailman/listinfo/vfio-users
Re: [vfio-users] intel_iommu=on and aacraid / Adaptec 3805
On Sun, 10 Jul 2016 13:17:41 -0500 Davidwrote: > I have run into a strange problem, while setting up my Fedora 24 box for > KVM, i noticed that my raid array stopped showing as an available drive. > After a lot of troubleshooting and reinstalling linux to this PC 2 more > times, i have narrowed it down to one setting in my GRUB config. > intel_iommu=on > > Just taking that one setting out and rebuilding my grub2-efi.cfg will make > the array readable again. When IOMMU is on, the system can see that there > is a raid card and array, but it lists the partition table as unknown. It > also will not successfully create a new GPT partition table on the array. > With IOMMU off, it can read the partition table and partition fine, read > and write data, and everything works fine. > > My raid is 4 1TB disks in Raid 10e, GPT partition formatted NTFS. > > $ lspci -v -s 03:0e.0 > 03:0e.0 RAID bus controller: Adaptec AAC-RAID > Subsystem: Adaptec 3805 > Flags: bus master, stepping, 66MHz, medium devsel, latency 32, IRQ 57 > Memory at fa60 (64-bit, non-prefetchable) [size=2M] > Expansion ROM at fa80 [disabled] [size=256K] > Capabilities: > Kernel driver in use: aacraid > Kernel modules: aacraid No DMAR faults in dmesg? There's a fair chance this is one of those devices that does DMA with the wrong requester ID and therefore enabling the IOMMU prevents it from working. We have a mechanism in the kernel to handle such DMA aliasing if you can verify the issue. Thanks, Alex ___ vfio-users mailing list vfio-users@redhat.com https://www.redhat.com/mailman/listinfo/vfio-users
Re: [vfio-users] Stability Testing/Benchmarking software for Windows VMs
It's been a long time since I did it, but I remember 3Dmark (commercial and free versions) having a looping mode, and I know Furmark was a real GPU killer last time I looked. If you have an eVGA graphics card, their Precision software includes a multiple-test-type version of Furmark built in. On Mon, Jul 11, 2016 at 9:19 PM, Brian Yglesias < br...@atlanticdigitalsolutions.com> wrote: > I'm not much of a gamer, so I'm looking for software to use for > troubleshooting stability issues. > > E.g. the overclockers all use prime95, or at least they used to. Is there > some analog de facto standard for GPU testing? > > The software I've found so far doesn't have a continuous testing mode. > Anything with diagnostic info would be a plus. > > ___ > vfio-users mailing list > vfio-users@redhat.com > https://www.redhat.com/mailman/listinfo/vfio-users > > ___ vfio-users mailing list vfio-users@redhat.com https://www.redhat.com/mailman/listinfo/vfio-users
Re: [vfio-users] intel_iommu=on and aacraid / Adaptec 3805
One can only imagine that the driver for this particular piece of hardware uses the iommu. The only other case I know of is the AMD cards with their proprietary driver, which is incompatible with activating iommu in the bios. Here the conflict must be in the aacraid driver. If I were smarter, I'd read the source of the driver. I suppose you could file a bug report. > On Jul 10, 2016, at 11:17 AM, Davidwrote: > > > I have run into a strange problem, while setting up my Fedora 24 box for KVM, > i noticed that my raid array stopped showing as an available drive. After a > lot of troubleshooting and reinstalling linux to this PC 2 more times, i have > narrowed it down to one setting in my GRUB config. intel_iommu=on > > Just taking that one setting out and rebuilding my grub2-efi.cfg will make > the array readable again. When IOMMU is on, the system can see that there is > a raid card and array, but it lists the partition table as unknown. It also > will not successfully create a new GPT partition table on the array. With > IOMMU off, it can read the partition table and partition fine, read and write > data, and everything works fine. > > My raid is 4 1TB disks in Raid 10e, GPT partition formatted NTFS. > > $ lspci -v -s 03:0e.0 > 03:0e.0 RAID bus controller: Adaptec AAC-RAID > Subsystem: Adaptec 3805 > Flags: bus master, stepping, 66MHz, medium devsel, latency 32, IRQ 57 > Memory at fa60 (64-bit, non-prefetchable) [size=2M] > Expansion ROM at fa80 [disabled] [size=256K] > Capabilities: > Kernel driver in use: aacraid > Kernel modules: aacraid > > $ lsblk -f > NAME FSTYPE LABEL UUID MOUNTPOINT > sda > ├─sda1 vfatefi 9E87-0ECF > ├─sda2 ext4boot 4f8d2664-8fc0-4779-ac18-6f0c06152d27 > └─sda3 crypto_LUKS ccee7bd2-4dce-4966-addb-bba6194ade93 > sdb > > * with IOMMU on, lsblk -f shows that it sees SDB, but with no file system or > label. > > $ uname -r > 4.6.3-300.fc24.x86_64 > > Screen shot of gnome Disks = > > > > > > -- > David > david...@gmail.com > ___ > vfio-users mailing list > vfio-users@redhat.com > https://www.redhat.com/mailman/listinfo/vfio-users ___ vfio-users mailing list vfio-users@redhat.com https://www.redhat.com/mailman/listinfo/vfio-users
Re: [vfio-users] GPU passthrough with Intel KVMGT, Nvidia card gets subsystem id 00000000 when passed through
Hi János, Unfortunately I'm not sure what the next step would be at this point. I suspect the VBIOS dump you got from Nouveau may not be correct, but this is little more than a guess. Given that the VBIOS is part of the main BIOS, you may need to extract it in some alternate way, however I can't say I know how to go about doing this. Hopefully someone else on the list will be able to provide a more useful response. Kind regards, Jack On 12/07/16 10:47, János Horváth wrote: > Hi Jack, > > Thanks for your reply. > > I have a native Windows 8.1 64bit system in dual boot currently, so I > didn't need the bootable USB trick. > The problem is that I have already tried dumping the VBIOS with GPU-Z > and it has failed. Apparently this GTX860M Maxwell GPU is somewhat > different and GPU-Z can't save the VBIOS. I have also read that the > VBIOS in this case is not in the card itself, rather it is included in > the main BIOS of the laptop. > > Regards, > János > > On Tue, Jul 12, 2016 at 2:40 AM Jack Coulterwrote: > > Hi János, > > I had problems myself when trying to extract a card's firmware via > Nouveau - the resulting VBIOS images gave me the same issue in the > VM - > a subsystem ID of all zeros. On a hunch, I made a bootable Windows USB > and dumped the VBIOS with GPU-Z - this resulting binary then > worked with > VFIO. You could give this a try and see if it resolves your issue. You > can use rufus (https://github.com/pbatard/rufus) to create a live > Windows environment ("Windows to Go" is the option you're after). > > Curious to hear how this goes, as I've recently acquired an Optimus > system myself and have been wondering if KVMGT + VFIO would finally > allow for Optimus to work in a VM. > > > Kind regards, > Jack > > On 11/07/16 06:26, János Horváth wrote: > > Hi all, > > > > I'm trying to get native Windows gaming performance while running > > Linux on my Optimus laptop (Lenovo Y50). > > Hardware: > > CPU: Intel Core i7-4710HQ > > RAM: 16GB DDR3 1600MHz > > GPU: Nvidia GeForce GTX860M & Intel HD 4600 > > It supports IOMMU and all the requirements for the pci passthrough. > > > > I have found a project by Intel, KVMGT > > > > (https://01.org/igvt-g/blogs/wangbo85/2016/intel-gvt-g-kvmgt-public-release-q12016) > > which lets me share the Intel Integrated Graphics (HD 4600) between > > the host and the guest, so I thought that I could replicate the > > Optimus feature inside the VM, while the host would be running > on the > > Intel HD 4600. I installed Intel's test release to the laptop > (Ubuntu > > 14.04 64bit with kernel 4.3.0 patched with support for KVMGT, and a > > patched QEMU and SeaBios) and created a Windows 8.1 64bit VM. > > It works great, the VM recognizes the Intel HD 4600 and I am able to > > choose which machine's (host or guest) screen to show on my display. > > (By the way this (Intel's) solution is by far the best for this > > purpose as I am able to switch between which OS to show on the > > displays not having to only rely on which monitor is plugged to > which > > output) > > The problem is (as expected) with the Nvidia GTX860M. I am able to > > pass through the device, but in Windows Device Manager, it shows the > > vendor and device ID correctly, but the subsystem shows as , > > so I am of course unable to install any driver. > > I have tried to use the vBIOS of the Nvidia card with the romfile > > option of QEMU, but it does not help (I extracted the vBIOS, > which is > > not UEFI capable btw, with nouveau driver from a live system). > > > > Here is the command I use to start QEMU: > > sudo /usr/bin/qemu-system-x86_64 -m 2048 -smp 2 -M pc -name kvmgt > > -drive > > > > file=/media/hjanos/Hjanos2TB/VM/GPU/GPU-intel.img,format=raw,index=0,media=disk > > -cdrom > > > /media/hjanos/Hjanos2TB/ISO/en_windows_8.1_with_update_x64_dvd_4065090.iso > > -bios /usr/bin/bios.bin -enable-kvm -vgt -vga vgt -display sdl > > -machine kernel_irqchip=on -vgt_high_gm_sz 384 -vgt_low_gm_sz 128 > > -vgt_fence_sz 4 -cpu > > > > host,kvm=off,hv_time,hv_relaxed,hv_vapic,hv_spinlocks=0x1fff,hv_vendor_id=Nvidia43FIX > > -usb -usbdevice tablet -net nic -net > tap,script=/home/hjanos/qemu-ifup > > -net user,smb=/home/hjanos -device > > > vfio-pci,host=01:00.0,romfile=/media/hjanos/Hjanos2TB/VM/GPU/vbios.rom > > > > I need help to figure out why does the subsystem id gets changed to > > when I pass through the Nvidia card, as I beleive that > makes > > the driver unable to install. (it is not code 43 as usual, rather it > > is code 28) > > > > Please let me know if you need any additional details. > > > > Thanks, > > Janos > > > > >