Re: [Qemu-devel] [RFC v2 0/8] KVM PCI/MSI passthrough with mach-virt
Tested-by: Yehuda Yitschak Tested on Armada-7040 using an intel IXGBE (82599ES). > -Original Message- > From: qemu-devel-bounces+yehuday=marvell@nongnu.org > [mailto:qemu-devel-bounces+yehuday=marvell@nongnu.org] On > Behalf Of Eric Auger > Sent: Friday, January 29, 2016 18:54 > To: eric.au...@st.com; eric.au...@linaro.org; qemu-devel@nongnu.org; > qemu-...@nongnu.org; peter.mayd...@linaro.org; > alex.william...@redhat.com; pranav.sawargaon...@gmail.com; > p.fe...@samsung.com; pbonz...@redhat.com; ag...@suse.de > Cc: bharat.bhus...@freescale.com; suravee.suthikulpa...@amd.com; > christoffer.d...@linaro.org > Subject: [Qemu-devel] [RFC v2 0/8] KVM PCI/MSI passthrough with mach- > virt > > This series enables KVM PCI/MSI passthrough with mach-virt. > > A new memory region type is introduced (reserved iova). On > vfio_listener_region_add this IOVA region is registered to the kernel with > VFIO_IOMMU_MAP_DMA (using the new > VFIO_DMA_MAP_FLAG_MSI_RESERVED_IOVA flag). > > The host VFIO PCI driver then can use this IOVA window to map some host > physical addresses, accessed by passthrough'ed PCI devices, through the > IOMMU. > The first goal is to map host MSI controller frames (GICv2M, > GITS_TRANSLATER). > > mach-virt currently instantiates a 16x64kB reserved IOVA window. This > provisions for future usage. Most probably this exceeds MSI binding needs. > To avoid wasting guest PA, we now map the reserved region onto the > platform bus MMIO. > > The series includes Pranav/Tushar' series: > QEMU, [v2 0/2] Generic PCIe host bridge INTx determination for INTx routing > ((https://lists.nongnu.org/archive/html/qemu-devel/2015- > 04/msg04361.html)) > > Those patches are not mandated for PCI/MSI passthrough to work but > without those, the following warning is observed and can puzzle the end- > user: > "qemu-system-aarch64: PCI: Bug - unimplemented PCI INTx routing (gpex- > pcihost)" > > Best Regards > > Eric > > Dependencies: > The series depends on kernel series: "[PATCH 00/10] KVM PCIe/MSI > passthrough on ARM/ARM64", (https://lkml.org/lkml/2016/1/26/371) > > Git: > QEMU: > https://git.linaro.org/people/eric.auger/qemu.git/shortlog/refs/heads/v2.5. > 0-pci-passthrough-rfc-v2 > > Kernel: > https://git.linaro.org/people/eric.auger/linux.git/shortlog/refs/heads/v4.5- > rc1-pcie-passthrough-v1 > > Testing: > - on ARM64 AMD Overdrive HW with one e1000e PCIe card. > > History: > RFC v1 -> RFC v2: > - now uses platform bus MMIO for mapping reserved IOVA region; hence > the > new patch file: > "hw: platform-bus: enable to map any memory region onto the platform- > bus" > > Eric Auger (8): > linux-headers: partial update for VFIO reserved IOVA registration > Add a function to determine interrupt number for INTx routing > Generic PCIe host bridge INTx determination for INTx routing > hw: vfio: common: introduce vfio_register_reserved_iova > memory: add reserved_iova region type > hw: platform-bus: enable to map any memory region onto the > platform-bus > hw: arm: virt: register reserved IOVA region > hw: vfio: common: adapt vfio_listeners for reserved_iova region > > hw/arm/virt.c | 23 > hw/core/platform-bus.c | 26 +++--- > hw/pci-host/gpex.c | 12 > hw/vfio/common.c | 68 > -- > include/exec/memory.h | 29 > include/hw/pci-host/gpex.h | 1 + > include/hw/platform-bus.h | 7 + > linux-headers/linux/vfio.h | 15 -- > memory.c | 11 > 9 files changed, 160 insertions(+), 32 deletions(-) > > -- > 1.9.1 >
Re: [Qemu-devel] [RFC v2 0/8] KVM PCI/MSI passthrough with mach-virt
On 29 January 2016 at 16:53, Eric Auger wrote: > This series enables KVM PCI/MSI passthrough with mach-virt. > > A new memory region type is introduced (reserved iova). On > vfio_listener_region_add this IOVA region is registered to the kernel with > VFIO_IOMMU_MAP_DMA (using the new VFIO_DMA_MAP_FLAG_MSI_RESERVED_IOVA flag). > > The host VFIO PCI driver then can use this IOVA window to map some host > physical addresses, accessed by passthrough'ed PCI devices, through the IOMMU. > The first goal is to map host MSI controller frames (GICv2M, GITS_TRANSLATER). > > mach-virt currently instantiates a 16x64kB reserved IOVA window. This > provisions for future usage. Most probably this exceeds MSI binding needs. > To avoid wasting guest PA, we now map the reserved region onto the > platform bus MMIO. > > The series includes Pranav/Tushar' series: > QEMU, [v2 0/2] Generic PCIe host bridge INTx determination for INTx routing > ((https://lists.nongnu.org/archive/html/qemu-devel/2015-04/msg04361.html)) > > Those patches are not mandated for PCI/MSI passthrough to work but without > those, the following warning is observed and can puzzle the end-user: > "qemu-system-aarch64: PCI: Bug - unimplemented PCI INTx routing > (gpex-pcihost)" I've replied with comments about the parts I care about; I'm leaving the rest of the review to others (VFIO related, etc). thanks -- PMM
[Qemu-devel] [RFC v2 0/8] KVM PCI/MSI passthrough with mach-virt
This series enables KVM PCI/MSI passthrough with mach-virt. A new memory region type is introduced (reserved iova). On vfio_listener_region_add this IOVA region is registered to the kernel with VFIO_IOMMU_MAP_DMA (using the new VFIO_DMA_MAP_FLAG_MSI_RESERVED_IOVA flag). The host VFIO PCI driver then can use this IOVA window to map some host physical addresses, accessed by passthrough'ed PCI devices, through the IOMMU. The first goal is to map host MSI controller frames (GICv2M, GITS_TRANSLATER). mach-virt currently instantiates a 16x64kB reserved IOVA window. This provisions for future usage. Most probably this exceeds MSI binding needs. To avoid wasting guest PA, we now map the reserved region onto the platform bus MMIO. The series includes Pranav/Tushar' series: QEMU, [v2 0/2] Generic PCIe host bridge INTx determination for INTx routing ((https://lists.nongnu.org/archive/html/qemu-devel/2015-04/msg04361.html)) Those patches are not mandated for PCI/MSI passthrough to work but without those, the following warning is observed and can puzzle the end-user: "qemu-system-aarch64: PCI: Bug - unimplemented PCI INTx routing (gpex-pcihost)" Best Regards Eric Dependencies: The series depends on kernel series: "[PATCH 00/10] KVM PCIe/MSI passthrough on ARM/ARM64", (https://lkml.org/lkml/2016/1/26/371) Git: QEMU: https://git.linaro.org/people/eric.auger/qemu.git/shortlog/refs/heads/v2.5.0-pci-passthrough-rfc-v2 Kernel: https://git.linaro.org/people/eric.auger/linux.git/shortlog/refs/heads/v4.5-rc1-pcie-passthrough-v1 Testing: - on ARM64 AMD Overdrive HW with one e1000e PCIe card. History: RFC v1 -> RFC v2: - now uses platform bus MMIO for mapping reserved IOVA region; hence the new patch file: "hw: platform-bus: enable to map any memory region onto the platform-bus" Eric Auger (8): linux-headers: partial update for VFIO reserved IOVA registration Add a function to determine interrupt number for INTx routing Generic PCIe host bridge INTx determination for INTx routing hw: vfio: common: introduce vfio_register_reserved_iova memory: add reserved_iova region type hw: platform-bus: enable to map any memory region onto the platform-bus hw: arm: virt: register reserved IOVA region hw: vfio: common: adapt vfio_listeners for reserved_iova region hw/arm/virt.c | 23 hw/core/platform-bus.c | 26 +++--- hw/pci-host/gpex.c | 12 hw/vfio/common.c | 68 -- include/exec/memory.h | 29 include/hw/pci-host/gpex.h | 1 + include/hw/platform-bus.h | 7 + linux-headers/linux/vfio.h | 15 -- memory.c | 11 9 files changed, 160 insertions(+), 32 deletions(-) -- 1.9.1